CA2229630A1 - Method and apparatus for annotation of medical imagery to facilitate patient identification, diagnosis, and treatment - Google Patents

Abstract

A method and apparatus for annotation of medical imagery to facilitate patient identification. diagnosis and treatment is characterized by an imaging device for producing a first signal representative of sensed characteristics of the individual and a minutiae generator which receives the first signal and produces a second signal representative of minutiae of the individual. the minutiae corresponding to specific branch points of blood vessels of the individual. A minutiae data generator analyzes the characteristics of minutiae and produces a third signal representative of the characteristics which is stored in a minutiae database for each of the plurality of known individuals and their medical conditions. The minutiae and minutiae data may be used to annotate medical imagery to facilitate subsequent image comparison byproviding standardized registration points and time-varying characteristics. A
minutiae matcher pairs corresponding second signals and third signals from a current patient with those from a database record, and the paired signals are used to align the images and compare them. The minutiae analysis techniques of the invention can be used to identify medical patients, assist in the diagnosis of medical conditions and detect and monitor the use of alcohol and drugs including anesthesia.

Description

~lETHOD AND APP.~R~TUS FOR ANNOTAT~ON OF MEDICAL
li~lAGERY TO FACILITATE P~TIENT IDENTIFICATION. DI~GNOSIS.
AND TREATi~IENT
This application is based on provisional applications No. 60!027.777 filed July ~9, 1996 No. 60i0~8 385 filed October 1~. 1996 and No. 60/078,387 filed October 15. 1996.

BACKGRO~TND OF THE rN~/~NTIO~
This invention relates generally to the field of image recognition and processing and specifically to methods and systems for identif~ing diagnosing, and treating people based on thermal minutiae ~ithin a person s body, primarily the fac e.
Improved methods for automated access control and surveillance are vital to ensure the continued security of nuclear ~veapon storage facilities as well as other sensitive or valuable items. Potential threats range from terrorist bombings, 1 ~ insider thefts. and industrial espionage to sabotage bv environmental activists.
There is concern for increased vigilance in the protection of critical strategic assets.
Current technologv being used for access control is not sufficient reliable secure fast rugged or cost effective for routine unattended operations at high-security locations. The challenge is to develop systems to secure facilities and~0 personnel from internal and e~ternal threats in a cost effective and timely manner.
Replacing human guards with automated systems can provide a significant cost savlngs.
The requirement to positivelv identifv each individual seel~ing access to a facility or to information or services is widespread. Manpower-intensive guard '5 brigades are deployed at public functions to protect celebrities and at locations where valuable or important items are stored. Guards are used to screen entrantsbased upon reco<Jnizin(J either the person or some credential he carries.
Identification credentials such as photo ID bad<~es and driver's licenses are ~.idel~
usecl for manual identification when cashin~J checl~s or usin(J credit cards. ~Ianual S chec:kin(7 of such identification cards mav not reco(Jnize cases where the card is forl1er~ or where the person usin(J it is not the ri(~htful owner of the card. To assist in solving that problem. more sophisticated identifvin~J characteristics mav be used on the card, and features ma~ be added to mal;e the card more difficult to coul?terfeit. The use of biometric characteristics such as finlJerprints. sionatures.
visual descriptions, or photo~Jraphs is becomin(J more common. Such information can either be readable manuall~ or encoded for readino bv an automated s!!stem.
When the identification system is fullv automated, without a human attendant. biometric sensors at the access location can compare the characteristics of a person at the location with the stored characteristics of the person he is ! 5 claimin(J to be. When initiallv issuin~J permission for a person to access abiornetrically-controlled s~ stem or location. his biometric characteristics arerecorded in the s~stem memorv. and also recorded on an identification card. for later comparison bv the s! stem controller witll his li~ e characteristics.

BRl:EF DESCRIPTIO!~ OF THE PRIOR .~RT
Current biometric identification s~stems include use of inkless fin~erprint s~vstems (called ~ live scan ' units), retinal scanners. hand (Jeometrv measurin( deviices, voice recoonition handwritino reco~nition. and facial recoonition sl;stems ~vhich use either visual or infrared cameras. Use of fin~erprints is (Jenerall~
considered the most secure method ior positive identification. Ho~ve~er. when used in an unattended mode~ fin(Jerprints can be lifted from one location or surface and positioned at another location. Therefore unattended use of fingerprints foridentification at locations requirimJ very hi~lh security is not acceptable. A more common limitation to widespread use of fin(rerprints for identification is the requirement for placin~ one or more clean fincTers on a ( lass plate for ima(~in(T bv S the fin(Terprint reco(Jnition system. This requires that the hands be free andre'latively clean. and that the (~lass plates be maintained intact and clean. The plates are vulnerable to vandalism. When used for access control at a busv location, there is a time delay associated with unloadin(T the hands and positionin(T the fin(Ters pro;perly. Also, users must cooperate with the s!~stem. In certain scenarios of use, 10 cooperation of the subject may be difficult to obtain. Furthermore, many persons have a reluctance to bein~T fin(Terprinted for an identity card. since they associate the process ~~ith criminal activities.
Fingerprints traditionall- have been the sole means of positive identification admissible as evidence in criminal trials in the U.S. Fin(Terprintin(T of criminals, 15 military personnel, persons seekino securit! clearances. and persons applyin(T for sensitive jobs has been performed f'or man-~ years. The FBI established and maintained a card file in which eacl- person s fin(Terprints w ere printed by rollino th finoers first on an ini;ed pad and then on the card. ~luch of the ori~Tinal FBI
fin~Terprint file of rolled prints has now been di(Jitized and made available on-line 20 for computer access. The process of di(fitizin~ the historical files. and theccntinuin~T task of maintainin(J current fin(Terprint files. has cost hundreds of millions of doilars durin(r the past ten vears alone. Aside from the labor costs of performino the di_itization and mana(TincT the search tasks throu(Th the database, si(~nificant R~D has been performed to develop specialized software for comparin(J
25 unknown fin ~erprints a ~ainst the database within a reasonable period of time. and specialized hardware has been developed to provide rapid response.

Inl;less technigues are now ~enerally used to produce a tenprint' card which substitutes for the former rolled print card. Common inkless techniques utilize polarized li~ht to illuminate the fin~ers. and li~Jht sensors to ima(~e the li~ht reflected and refracted t'rom the rid_es. The resultin_ ima~re can be more 5 consistent and hio~her qualih~ than the rolled prints since inconsistencies in the amount of in'~; applied and in the pressure used to transfer the print to paper are not a factor.
Automated fin_erprint matchin(!~ techniques have been developed which rapidly classify an unlinown print and then search throu_h the portion of the 10 database associated ~ ith that class lookincr for a match. Unknown prints may be from a tenprint'' card or may be latent prints which have been lifted from a crime scene. A latent print mav include a sizable area of one or more fin<Jers such as on a water glass or it may include only a portion of one or more finoers such as on a telephone keypad. Latent prints may be found on top of other latent prints, such as 15 when several people have used the same telephone.
Matchincr techniques often e~tract minutiae points from the prints and then compare the sets of minutiae rather than compare entire prints. ~ arious classifications of minutiae types have been proposed by different companies and aul:horities. .~n e~ample is _iven from the Costello U.S. Patent ~o. 4.947~443. Si~
~0 types of characteristic features'' are presented in this patent, each one relatin~r to a type of minutia. This fin(Jerprint matchin_ technique references the tvpe, orientation, and location of each characteristic and each and every other characteristic. Usin_ this approach~ on the order of 80 to 150 minutia points are identified in each fin(rerprint. Other fin_erprint minutiae e~traction and matchin ~5 pal:ents produce essentially the same number of minutiae~ ~vith difference in what fea.tures of the set of minutiae are considered in attempted matchin_ and in how the matchin(~ is performed. In U.S. courts. evidential~ rules have traditionallv required thiat 16 or more minutia points be found to correspond between two prints in order for them to be considered to be from the same person. The determination of likelv matchin(J prints is oenerally assisted or performed entirely by a computer svstem;
S however, the final decree of a match is made bv a fin(Jerprint e~pert, who re~iews the computer system results.
Matches bet-~ een different prints taken from the same fin(~er are never perfect, since the fin(~ers are deformable, three-dimensional, connected and jointed structures which leave two-dimensional prints on surfaces they encounter throu~h10 pressure. The exact an(~les between the finoers and the surfaces. the amount and direction of pressure. and the effect of movement between the fin(~ers and the su-rfaces all cause variations in the exact prints produced. Even when prints are produced by a live scan technique, variation in the li(l~hting hand position, oil or dust on the finc~ers. use of lotions. and scratches or paper cuts will produce minor 15 variations in the prints produced.
Therefore. the exact number, position and characteristics of minutiae extracted from two prints mav be different even thou~h thev are produced by the same fincJer. The challenoe for an automated fin(Jerprint identification system is to recoC~nize allowable minor variations in actual matchin(J prints ~hile not allowing ~0 variations so wide that mismatches occur. Several .~FIS products are now commercially offered which provide acceptable accuracv. Local and re~ional police forces may use smaller databases which contain only the prints of personshistoricallv associated with their areas, rather than relyin~ on federal resources to search the entire nationwide FBI files. Smaller scale finc~erprint svstem. such as ~S those associated with a system which controls access to an office buildinc~, mav use the same minutiae matchin(~ techniques.

\~ith rolled and live-scan prints. the orientation of each print. and the fin(Jer to which it corresponds is known. Also. quality checks can be built into the process such that repeat prints mav be taken to insure qualitv when needed. In the case of latents. howe~er. the analysis is done after the fact. It is not known which finger left the print. and the orientation of the fin(rer mav be in doubt when only a pan:ial print is found. Therefore, matchino of latents is much more difficult than matchin~J of rolled or live scan prints.
Various minutiae extraction al(gjorithms are used in current fingerprint identification systems. some of which merely uti]ize the location of the minutia10 points and others of w hich utilize also additional information about the tvpe of minutia each point represents. For example. simple graph matching techniques canbe used to compare the follow-the-dots vectors ~Jenerated bv connecting the minutia points in order forced by considering intersections with a spiral from the centerpoint of the fingerprint. Alternately. the rid(~e angle at each minutia point ] j can be considered and matched along with the coordinates. in a best-fit attempt to match each unl~nown print to each known print. .~ measure of goodness of fit canthen be computed and used to rank other possible matches.
U.S. patent No. 4~5~5,859 to Bowles teaches a pattern reco~Jnition svstem which detects line bifurcations and line endin~rs. denoted minutiae. in a pattern of ~0 lines such as are found in a fingerprint. Accordin(r to this reference. the FBI uses an automatic fin~Jerprint identification svstem entitled F~NDER'' which uses an optical scan reader. The information is then enhanced to eliminate grays and fill in gaps in the ridges. A 16x16 increment square window scans the fingerprint, an increment being a tenth of a millimeter. Thus, a window advances throu~Jh the ~ ~5 fin~;erprint in increments of tenth of millimeter and lool;s for ridges which enter the window but do not exit it. When such a rid~e is identified. its coordinate ]ocation is s~:ored and the rid(re is analyzed to establish an an_le. theta. of the ridge at the terrnination. The data are then re-scanned to look for terminations of vallevs which are ridge bifurcations. The additional coordinates and anc~les of each of the inverted endin(~ points also are stored.
In latent prints the distances between rido~es of a fin_erprint averao~e 0.4 millimeters but can var~ b-~ a factor of ~ for an~ individual fin_er depending on sl;in displacement when the fin_er contacts the hard surface normally encountered in establishin_ a print.
~ ~ known al(~orithm of the National Institute for Standard Technolo(~v can be used to compare a previously stored electronic image of minutiae coordinate locations with the minutiae locations identified and stored by the computer.
U.S. Patent No. 5,040.~4 to Hara discloses a fingerprint processing s~ stem capable of detectin_ a core of a fingerprint image by statisticallv processing parameters. Hara's invention provides a system to determine a core in the I 5 fin~erprint image and/or to detect directions and curvatures of rid~es of the fin(~erprint imag~e prior to detection of the position of the core. This reference defines minutiae as abrupt endings. bifurcations. and branches.
I~.S. Patent No. 4,790,564 to Larcher teaches a process and apparatus for matching fingerprints based upon comparing the minutiae of each print in a ~0 database with precomputed vector ima-~es of search minutiae in a search print to be identified, comparing position and angle. a result of such comparison being a matching score indicating the probability of a match between the angle of a filepri:nt minutiae and the ancle of precomputed vector ima(~es of the search minutiae.
Over or under-inking of a rolled print can change the apparent type of minutiae ~5 associated with a particular point from one printin(~ to the ne~t. However, not all colTespondin_ minutiae will appear to chan_e type in the two prints. Therefore.

matching for type as ~ ell as for x and y coordinates provides a stricter match requirement and results in better system accuracy. Larcher assigns higher values to minutiae whicll match in ~;.y and type.
As Larcher points out~ there are ad~,antages to matching minutiae rather than the entire image of the fin~erprint in itself. An elementary matching operation cornprises the comparison of two sets of minutiae~ i.e.~ two sets of points~ each point having three coordinates x, y, and a. An elementar matcher attempts to superimpose the two sets of points? in order to count the number of minutiae which are common to the two fingerprints Numerous other schemes for matching fin~-~erprints are known. For example, matchers referred to in Wegstein Technical Note 538 ofthe National Bureau of Standards (1970) as Ml9, M27~ and M3~, determine whether two fingerprints come from the same finger b~ computing the densitv of clusters of points in Dx-Dv space wllere Dx and Dy are the respective differences in x and y15 coordinates for the minutiae of two fingerprints. Experimental results referred to in this reference indicate that in Dx-Dy space pohlts tend to be located at random when coming from different fingerprints, uhereas points tend to form a cluster when coming~ from finoerprints from the same fino~er.
In the Ml9 matcher~ the assumption is made that the transformation needed 20 to superimpose the two sets of minutiae points is a translation onlv. The M27mal:cher is an M l 9 matcher ~~ ith a new scoring function intended to take intoaccount greater translation displacements. The M32 matcher takes into account small rotations between two fingerprints in the followin(J way: first an l\I27 malcher comparison is made between the two fingerprints; then~ one ofthe two 25 prints is rotated throu_h - V'' degrees from its origillal position and a new 1\127 comparison is made. .~ll to_ether an M32 matcher operation consists of seven M27 comparisons. correspondin~ to the followin_ values for the an_le ~, i.e.
V= -15, -10 -5~ 0 +j +10, +15 decrees.
Minutiae mav be e~ctracted manuallv or automatically. Automatic systems g enerally require better quality ima_ery. The matcher en_ine must allow for some de ~ree of inaccuracv or ~ ariabilih~ ~.;ith respect to each of the encoded coordinates due to human operator bias or precision limitations of automated feature extraction processes.
Larcher disclosed the use and comparison of type of minutiae, since there is a _reater match accuracy when ridge endings are compared to ridge endings, and 10 bifi~rcations to bifurcations, as opposed to comparing one ridge ending to one bifi~rcation.
Other known approaches compare two sets of ima(?~e features points to det:ermine if they are from two similar objects as disclosed for e~ample in Sclaroff and Pentland, MIT Media Laboratory, Perceptual Computing Technical Report 15 ~¢304. This reference sug( ests that first a body-centered coordinate frame be de~ermined for each object, and then an attempt be made to match up the feature pomts.
Many methods of finding a bodv-centered frame ha~;e been su_gested.
including moment of inertia methods~ symmetry finders~ and polar Fourier ~0 descriptors. These methods àenerally suffer from three difficulties: sampling error, parameterization error, and non-uniqueness The technique used in Sclaroffand Pe-ntman disclosure has the limitation that it cannot reliably match largely occluded or partial objects.
Known techniques associated with fin_erprint minutiae e~;traction and 25 matchino~ can be summarized as follows First, an unknown fin_er is scanned opticallv;

Second. the ima~Je is divided into pi~cels. w here the size of the pixel relatesto the quality of the result desired;

Third. certain pixels are selected as minutiae points:

Fourth. each minutia is assit~ned a vector havin(J ma(~nitude and directional information in relation to the surrounding characteristics of the fin(Jerprint.
Typically for each fin(Jerprint. there would be a substantial number of minutia vectors characterizin~ its image:

Fifth, the set of minutia vectors of the unl;nown print are compared by computer to the set of vectors of known prints; and Sixth. the comparison results are used to select potential matches and provide a (roodness of fit indication between the unknown and known prints.

Numerous approaches to reco~nition usin~J visible li~ht ima(Jin~J of faces have been proposed. ~Iany of them apply standard pattern matchino techniques:
others involve definition of face metrics.
U.S. patent ~io. 4.975.969 to Tal discloses a method and apparatus for uniquely identifvin(~ individuals by measurement of particular physical characteristics viewable by the naked eye or b,v ima~in(J in the visible spectrum.
This reference defined facial parameters which are the distances bet~veen identifiable parameters on the human face. and/or ratios of the facial parameters, ~0 and teaches that the~ can be used to identify an indi~,-idual since the set of parameters for each individual is unique.

Tal's approach utilizes visible features on the face~ and therefore cannot be relied upon to distin_uish behveen faces ha~in ~ similar visual t'eatures~ for example as would be the case with identical twins. In addition, the 'rubber sheetin_'' effect caused by changes in facial e~pression, the a_in_ effects which cause leng~tllenin(~g S o:Fthe nose~ thinnin_ of the lips, wrinl;les, and deepening of the creases on the sides or~the nose, all cause chang~es in the parameters and in ratios relied on in this method. Furthermore the parameters and ratios of anv particular person's face may be measured by anvone tal~ing a photograph, and thereby used to select or disc~uise another person to appear to be that person. Therefore, the security 10 provided bv such a technique may not be adequate for unattended or highly .
sensltlve locatlons.
Still another l~nown scheme utilizes ei ~enanalvsis of visual face ima_es to develop a set of characteristic features. Pentland~ ~ iew-Based and Modular Eic~enspaces for Face Recog~nition, MIT Media Laborator,v Perceptual Computin_ 15 Section, Technical Report No. 245 Faces are then described in terms of wei(Jhtin~
on those features. The approach claims to accommodate head position chan_es and the wearin ~ of_lasses~ as well as chan_es in t'acial e~pressions. This disclosure teaches that pre-processin ~ for re_istration is essential to ei_en~ector reco_~nition sy,tems. The processing required to establish the ei_en~ector set is e~tensive 20 especially for lar_e databases. Addition of new faces to the database requires the re running of the ei_enanalysis. According~ly, use of ei_enanalysis may not be appropriate for use in a ~eneral face identification system such as would be an,lloc~ous to the FBI's and AFIS finc~erprint svstem.
Visible metrics tvpically require _round truth distance measurements unless 25 the~v rely strictly upon ratios of measurements. Thus such systems can be fooled by intentional dis_uises, and they are subject to v ariations caused by facial e~pressions~ makeup~ sunbun1s~ shadows and similar unintentional dis_uises.
Del:ectin_ the wearing~ of dis_uises and distin_uishing between identical twins may be done from visible ima_ery if sufficient resolution and controlled lightin_ isavailable. However, that significantly increases the computational comple~ity of5 the identification tasl; and makes the reco_nition accuracy ~ulnerable to uni-ntentional normal variations.
From the standpoint of evidentiary use. it mi_ht also be argued that the application of eigenanalvsis to a very lar_e database of faces such as all mug shots in the FBI files would be considered so esoteric bv the public at lar_e that 10 automated matches based upon its use will not readilv be acceptable to a jury as convincing evidence of identity. Bv comparison. techniques based on minutiae mal:ching technique such as are used with fing~erprint identification, would be expected to find a more understanding reception bv the law enforcement cornmunity, and to be more acceptable for evidentiarv purposes within a reasonable 15 nurnber of years after their introduction.
One known scheme usin(~; facial thermo_rams for identification is described in the Prokoski et al ~1 S. Patent No. 5,163~094 wl1ich discloses definino elemental shapes'' in the surface thermal image produced by the underlying vascular structure of blood vessels beneath the s};in. Dependin_ on the environment of use~ thermal20 facial identification mav provide _reater securitv over identification from visual ima._es and ma~ therefore be considered preferable. It is e~tremely difficult, if not impossible, to counterfeit or forcge one face to look like another in infrared whereas it is often possible to disguise one person to look like another in visible li_ht. However. the use of elemental shapes is found in practice to be vulnerable to 25 such variables as head rotation and tilt, ambient and physiological temperature changes. variations in imag~in( and processing systems~ and distortions or obstructions in a facial image (e.g., due to eyeglasses).
t~crh~l Eigenanalysis of the elemental shapes of alfacial image has also been used ~~
for recognition. In one approach, several sets of elemental shapes are produced for each image by imposing different thermal banding constraints. The totality. of 5 shapes are then analyzed w ith respect to a library of facial thermal images.
Eioenshape analysis is used to compare the characteristics of shapes in each person's images. Eleven characteristics of each shape are considered, including:perimeter, area, centroid x and y locations, minimum and maximum chord length through the centroid, standard deviation of that length. minimum and maximum 10 chcrd length between perimeter points, standard deviation ofthat length, and area/perimeter.
Each person's image is then characterized by a set of I l-coefficient vectors.
The difference in eigenspace between any two images is calculated to yield a measurement to which a threshold was applied to make a match/no match"
15 decision. In practice, such a system yields a useful method and apparatus for some applications. However, the calculation techniques for such a system are computationally intensive and require additional computational analysis of the entire database when new images are added. As with others of the prior known techniques, recognition is seriously impacted by edge effects due to head rotation 20 and tilt, and by loss of definition in very cold or very hot faces.
None of the known techniques for facial analysis is believed to be sufficiently robust and computationally straightfonA~ard to allow practical application of such a scheme for highly sensitive unattended security applications.
Therefore, the need remains for a system and method that can be used to ~ 25 reliably recognize and verify the identity of an imaged person without manual ~ assistance and without cooperation from the person being identified.

SU~ RY OF THE IN~ ~rTIO~
In accordance witll the present invention. a system for recocmizing~ faces comprises a thermal imaoinCJ device, a minutiae generator, a minutiae data generator, and a minutiae matcher. The thermal imagin(J device produces a signal5 representative of tlle thermal characteristics of a new face The minutiae generator is c onnected to the thermal imaging device and produces a signal representative of thermal facial minutiae of the new face The minutiae data _enerator stores minutiae data corresponding to known faces The minutiae matcher is connected to the minutiae ~renerator and the minutiae data oenerator and compares minutiae10 of the new face and of the known faces. producing a silJnal representative of a match between the new face and one of the old faces In another aspect of the invention. a method of reco ~nizing faces senses thermal characteristics of known faces, identifies minutiae of the known faces.
senses thermal characteristics of a new face. identifies minutia of the new face.
15 determines a distance metric from each of the knov.;n faces to the new face and determines a match between the new face and one of the o]d faces based on the distance metrics.
In still another aspect of the invention. faces are classified according to thermal minutiae. and facial minutiae data are encoded as a number of bits by 20 overlavino a grid of cells on a thermal representation of face, setting a bit to a first stal:e if any minutiae are located within the cell corresponding to that bit. and setl:ino the bit to a second state if none of the minutiae are located witllin the cell corresponding to that bit.
In yee further aspects of the invention. other imagino modalities are used.
25 and other body parts or objects are used. for minutiae-based reco(Jnition Techniques for identif ing medical patients. diagnosin(~ medical conditions.

identifying drug and alcohol users, and assisting with the positioning of surgical instruments are also achieved with the present invention.

BR~EF DESCRIPTION OF THE FIGURES
Other objects and advantages of the invention will become apparent from a 5 stud.y of the following specification, when viewed in the light of the accompanying drawing, in which:

Fig. I is a block diagram of a system for recognizing faces according to the present invention;

Figs. 2a and 2b are front and side views respectively? of the 10vascular system of the human head, with the location of thermal minutiae being indicated in Fig. 2b;

5c~eh~c Fig. 2c is a ~ view of the vascular svstem of the human brain, Fig. 3 is a front view of the human vascular system illustrating the location of thermal minutiae therein.

15Fig. 4 illustrates a contour plot derived from a facial thermogram and identified facial features in accordance with the invention;

Fig. 5 illustrates a contour plot derived from a facial thermooram and minutia points in accordance with the invention;

Fig~. 6 illustrates a =rid of cells overlaid on a contour plot derived from a facial thermo_ram in accordance ~ ith the in~ ention:

Fj(J. 7 is facial thermogram taken from a distance of approximately fifteen feet in accordance with the invention.

Fio. 8 is a block dia=ram of the apparatus for annotatino an image of the human body according to the invention~
., Fig. 9 is a block diaoram of a modified apparatus of Fio. 8 for identifyino reference minutiae in an annotated image;

Figs. IOa. IOb, and lOc illustrate facial minutiae superimposed on three different thermal imaoes, respectively. of the same face:

Fig. I 1 is a block diagram of an apparatus for annotatingJ a medical ima_e of the human bodv according to the invention:

Fio. I 2 is a thermooram of the upper chest area of an indi~ idual taken from a distance of approximatelv fifteen feet in accordance ~~-ith the I j invention:

Fig. I ~ illustrates the correspondin_ anatomv for the thermogram of Fig. 12;

Fi~r. 1~ is a block dia(Jram representin(J apparatus for maintainino the position of a sur(Jica] instrument relative to a sur~Jical site accordin~J to the invention:

Fi~. 15 is a facial thermo ~ram of an alcohol-free individual:

Fi~. 16 is a facial thermo~ram of the individual of Fi~. I 5 under the influence of alcohol.

Fj'J I7jS a (Jraph representin~J the thermal si(Jnatures of selected minutia points of an individual prior and subsequent to use of alcohol:

Fi'J. 18 is a block dia_ram of the apparatus for detection of alcohol and dru(J use b~, an individual accordin_ to the invention;

Fi~J. 19 is a (Jraph representin~J the results of a statistical anal~ sis of dru~ users in a random population. and Fi~ 20 is a block diaoram of the apparatus for detection of alcohol and dru~ use in a random population DETAILED DESCRIPTION

Facial Minutiae E!~traction In Fig~ l . there is shown a svstem l 00 t'or personal identification in accordance with the present invention S~stem 100 includes seven major subivstems a thermal imaginQ device 102. minutiae o~enerator 104~ a minutiae data generator 105, minutiae matcher 106, minutiae database 111~ microprocessor 110, and output display/processin_ device 114 In a preferred embodiment. minutiae oenerator 104~ minutiae data _~enerator 105, and minutiae matcher 106 are all implemented by prooram instructions stored in a prog~ram memory 108~ and 10 pro~Jram memory 10'~ microprocessor 1 10. and minutiae database 1 1 1 are implemented bv a programmed conventional computer 1 17.
In operation. thermal imaoin_ device 10~ obtains a thermal ima~ge of the face of person 101 A di_ital signal representative of the thermal image is provided as h1put to minutiae cgenerator 104, which Qgenerates si=nals representative of 15 thermal facial minutiae points for 101 These minutiae points have a number ofcharacteristics includino a specific location ~~ ithin the person or relative to other min-utia, the apparent temperature at a Civen time the temperature sionature over a period of time. whether the minutia corresponds to a vein or arterv. the width of the blood vessel and the vector direction of branchino blood vessels from the ~0 minutia These and other characteristics are sensed and data relative thereto are gJenerated by a minutia data oenerator 105 This data is stored in the minutiae database 11 ] The minutiae matcher 106 compares minutiae data for known individuals which has been stored in the database with that for unknown individuals currently being ima_ed bv the imaging device 10~ If a match is detected. a correspondin(~ si(~nal is sent to the output displayiprocessin~ device 11~.
In a preferred embodiment, output display/processin(~ device 114 comprises circuitry to permit or deny access to a secured facility dependin~J on the results of the matching performed bv minutiae matcher 106. ~n one embodiment, access is 5 perrnitted if the person lO I is recoonized as one of a oroup of authorized personnel. In a second embodiment, access is denied if the person 101 is recoonized as one of a oroup of unauthorized personnel. In vet another embodiment, access is denied if the person lOI is not recoonized by system lO0.
System lO0 thus considers hidden micro parameter which lie below the sl;in 10 surface, and which cannot be easily for~ed. if at all. The lar~e number of such micro parameters considered renders it essentially impossible to search for a person to match another person s set of micro parameters. Furthermore. the particular infrared band used for imagin(7 by thermal imaginlJ device 102 mav be kept secret.
or multiple bands mav be used, which further increases the difficultv involved in con-lpromisino svstem 100. The underlvin(~ features détected by system 100 are essentially hardwired' into the face at bir~h and remain relatively unaffected b!~
a~Jin~. thus providin(r for less inherent variabilitv than found in known reco(Jnition sy,stems. Althouoh thermal facial minutiae have some aspects related to, and e~tractable from. elemental shapes and mav be ta~oed to reflect the elemental shape parameters (such as bv ta~gino with fractal dimensions), minutiae eYtraction does not require production or consideration of elemental shapes. Furthermore, the comparison of thermal facial minutiae is computationall!~ straightforward and introduces sionificantly less processing overhead than the known approaches usedfor lemplate or shape comparisons.
Thermal imaoin(J device 102 mav be any device that produces a sional representative of the thermal characteristics of the face of person l O I . In a ~0 preferred embodiment a conventional di~ital ~ideo camera sensitive to thermal eneroy is used for the thermal imagin~ device 102. As described herein? it is found that tractable imagerv for facial identification may be derived from passively obt,~ined infrared images of facial heat emanations which can be detected by commercially available thermal ima(~ino devices sensitive in the 3 to 12 micron wavelens~th band. Unlih-e fingerprints that are characterized bv a limited ran~e of intensity values corresponding to three dimensional ridgJes which are essentiallv concentric rings about a single center, plus anomalous arches line endings, and bifurcations, facial thermo~rams are aenerally characterized by continuously 10 var~'in~ wide distribution of temperatures includin~J multiple maxima and minima values. Where the skin surface is unbroken there is gradual variation of temperatures from the hot areas on either side of the nose to the relativelv cool areas of the ears and cheeks. The eyes appear to be cooler than the rest of the face. The nostrils and mouth, and surrounding areas~ will look v~,-arm or cool I 5 dependin(r upon whether the subject is inha]in~J or e~chaiing throuoh them.
Discontinuities in the skin surface temperature may be evident where scars, moles, burns. and areas of infection are found.
In some applications, thermal imaging device 102 mav be adapted for attended operation usin~J cooperative persons 101 and a human supervisor. as with 20 identification svstems based on rolled fingerprints. In these applications the supervisor can ensure that person l O l is properly positioned and can adjust gain focus, and other parameter ofthermal imagin(~ de~,ice 102 to optimize the qualitv ofthe thermal image produced by thermal imagJin~ device 102. In other applications, thermal imaging device 102 is adapted for unattended, stand-alone 75 operation, for instance with live scans used for access control to a remote secure facilitv. System 100 can further be configured based on an e~pectation that person 100 will be either cooperative (e.(J.. movin ~ to a specific requested location for optimal ima ~in_) or uncooperative (e.g.. a mere passer-by). In environments where uncooperative persons are e~pected. identification will be facilitated by col]ectin_~ the maximum possible amount of data. for instance b- usin_ multiple thermal ima(~rin J devices 102 and fast frame (i.e., samplin ~) rates. Additional related data. referred to herein as ~'2round truth'' data. may be collected as well to provide information on factors such as ambient temperature, absolute size of theimaged face, or the distance of the ima,~ed face from thermal imaoinC~ device 102.
Any portion of the body can be utilized for identification, but the face is prei'erred due to its typical accessibility for ima_in_. In Fi_. 2a there is represented the vascular system for a human face and in Fi ~. 2b there are shown minutiae pointsl50forthefaceofFi~.2a. InFicr 3 selectedminutiaepointslS0 thrc,u2hout a human body are shown.
Since parts of the face mav be bloci~ed by (~lasses facial hair. or orientation to thermal ima_ing device 10~. system 100 provides for identification based on partial faces. A sufficient number of minutiae may be obtainable from portions of the ~ace not blocked bv (~lasses, facial hair. or other concealment, to permit mat~-hin~. Aiternatelv if fewer than a minimum number of minutiae specified for a particular scenario are e~tracted by system 100 for a particuiar person 101 in an unattended settin ~. that person 101 may be considered by system 100 to be potentially dis ~uised. and output/display processino de~ice 11~ may cause an alarm to be ~g~enerated to alert ouard personnel to that possibility.
Various perturbations, such as facial e.Ypression chanlJes, can distort the relative locations of minutiae points. This is analo(~ous to the deformations that occur in fin_erprints due to movement or pressure between the fin(~ers and the prinl: surface. As described below, minutiae matcher 106 allows for some variations in the position and characteristics of the minutiae~ as well as in the subset of minutiae which are seen due to the field of view of thermal imaoin_ device 102 and. to possible obstruction of certain areas of the face in the ima_e.
As set forth in _reater detail herein, in one embodiment the minutiae dat.abase 112 is partitioned by classifyinu data correspondino, to faces based on minutiae-related characteristics as oenerated bv the minutiae data ~enerator 105.
In alternative embodiments, other characteristics mav be used for such classification. Such classification is found to reduce search requirements in connection with the operation of database 112 and minutiae matcher 106.

Minutiae Generator 104 In a preferred embodiment, minutiae oenerator 104 performs seven major functions: desi=nation of faces axes, testin-J of face axes validity; normalization;
production of thermal contour lines. establishment of threshold radius of cur~ ature;
selection of minutiae; and assionment of characteristics to minutiae. Each of these l j functions is described in oreater detail belo~ .

I. Desio,nation of Face ~xes Referrin(~ now also to Fio. 4, there is shown a facial thermoo,ram 200 as produced by thermal ima~in~ device 10~. The thermo~ram is in the form of a contour plot derived as set forth below. In a preferred embodiment, thermooram 200 produced b~ thermal ima~in_ device 102 is represented b! di_ital si~nals, but if an analoo, thermal ima Tino, device 102 is used, minutiae oenerator 104 can include conventional analoo-to di_ital conversion circuitry to obtain facial thermos~ram 200 as a digital signal representation of the face of person 101.
Once the facial thermogram 200is produced minutiae generator 104 locates a number of facial features on thermogram 200, either with manual assistance or automaticallv by using conventional techniques and structures as described herein: left and ri2ht canthi 201, 202 left and right nostrils 203, 204, ancl mouth area 205.
For clarity in illustration, only a single lcft canthus and ~1 singlc canthus arc 153~
~ shown in Fig. ~ . In a preferred embodiment, minutiae generator 104 locates the left canthi area and the right canthi area and determines the centroid for each area.
The location ofthe centroid is essentially independent ofthe grey scale allocation of th.e analog thermal imaoe produced by the camera 102. The centroids are referred to as left and right canthi 201, 202 herein.
Minutiae generator 104 also determines an eve line 206 between left and ri~ht canthi centroids 201, 202. This being done, minutiae ~generator determines a mouth line 205 parallel to eye line 206 and passin2 through the centroid of mouth area 205. Minutiae generator 104 next determines a vertical central line 208 perpendicular to eye line 206 and mouth area 205 and intersecting eye line 206 midvvay between left and right canthi centroids 201, 202. Minutiae oenerator 104then determines a face center point 209 on central line 208 midway between the poinl:s of intersection of vertical central line 208 with eye line 206 and mouth 205.
Minutiae generator 104 further determines a horizontal center line 210 perpendicular to the vertical central line 208 and passing through face center point 209. Vertical central line 208 and horizontal central line 210 are designated as face axes. Numerous other features may be used to define face axes but in general it is preferable to define face axes based on areas of the face that are not greatly ~ deformable.

.

2~
Other techniques may be used for location of the face center point 209 in those cases where the preferred use of facial svmmetry and recoonizable thermal features does not suffice. For example, other techniques mav be called for with respect to facial ima(res in which an eye pa~ch is worn, eye~Jlasses are not 5 symmetrical, onlv a partial face is imaoed. the lower face is covered. or the thermal pattern of the face is unusually distorted. The face center point 209 may in fact be outside of the boundaries of the facial ima=e~ for instance where onlv a partialfacial imaoe is obtained due to the face beino partially blocked by another face or some other object. If the person 101 is wearin~ glasses the pattern ofthe ~rlasses, 10 which typically block the infrared emissions from the face and thereby produce an extended cold area with sharp thermal discontinuity. can be used to determine approximate face a~es. Additional techniques include manual location of the facecenter point 209 and preprocessino usin~ known techniques to locate the approximate area of the face center point 209. As described below, the face a~;es 15 may be tested for validity to determine whether the ima(re requires anv such special treatment.

II. Testino the ~, alidity of Face Axes Since the known techniques for identifvino left and ri~rht canthi centroids 201, 202, left and rioht nostrils 20~, 204, and mouth area 205 are subject to 20 artifacts and other sources of error, and since some ima~,es of faces are si~nificantly asvmmetric or have features that are entirelv missin(J (e.~J., due to person 101wearin(r an eye patch or havino a disfioured face), minutiae oenerator 10~ performs checks to help spot instances where these points mav have been incorrectl~ located or where unusual facial imaoes are encountered. First? a check is made to ensure that vertical central line 208 and mouth line 207 intersect w ithin mouth area 205 Ne.Yt, a check is made to ensure that vertical central line 20S intersects a line connectin_~ left and ri ~ht nostrils 203.204 at point between left nostril 203 and ri(~ht nostril 204. If either of these conditions is not met. the face is considered to S be a special case callin_ for manual intervention to determine the best approximation for face axes III Normalization In practice, it is found that preprocessinQ through normalization of ima_e size provides advantages in later reco_nition Accordin ~Iv. minutiae C~enerator 104 uses the distances between left and ri_ht canthi centroids 201 and 202 and the distance from face center 209 to eye line 206 to compare the size of facial thel-mOC,ram 200 with a standard ima_e size In a preferred embodiment. Iinear correction in the vertical and horizontal dimensions is used to normalize the size of facial thermogram 200 to match the standard, but other normalization models 15 could be used as well, IV Production of Thermal Contour Lines As provided by thermal ima_in_ device 102. facial thermogram 200 consists onl~ of an ordered list of thermal values correspondin~-~ to each small portion of the ima~ed face Minutiae ~enerator 104 emplovs the followin(!~ procedure to produce 20 thermal contour lines for facial thermo ~ram 200:

'6 a. For a digitized image havin~J N bits of resolution. or ~' bands of therma] values determine thermal contour lines havin(J a particular current one of the ~' v alues.

b. Produce minutiae in accordance witll the steps below for the contour lines of the current v alue.

c. Repeat a and b above. each time using new one of the ~' values for the ' current" value. until the desired number of minutiae have been extracted of all of the possible values have been processed.

d. If the desired number of minutiae have not been extracted.
repeat the process be(~innin(~ ~ ith 2'-' bands of values. and reduce the number of bands by I ~-ith each iteration. skippin g those that are po--;ers of 2, until the desired number of minutiae have been extracted or until no further reduction in bands can be achieved.

I ~ Various other techniques for gJeneratin 7 contour lines may also be used.
wit:h the goal being obtainingJ a sufficientl- large number of minutiae for unique recognition, without producing too many spurious minutiae. Spurious minutiae increase processing overhead v~-ithout benefirting reco~Jnition. The number of thelmal bands that will produce an appropriate number of minutiae is readilv ~0 determined by trial and error for any particular application of system 100.

V. Establishment of ~laximum Radius of Curvature In a preferred embodiment, points on a thermal contour are considered minutiae if thev form inflection points for the contour. However, to avoid artifacts resulting in too many minutiae bein~ selected, only inflection points for curvesbelow a threshold radius will be considered minutiae. Therefore, minutiae generator 104 selects a maximum radius of curvature to be used in determining minutiae, based on characteristics of system 100 such as the resolution of thermal imaoin_ device 102. the lens used. the quality of the recording and processing system, the desired number of minutiae to be extracted, the desired sensitivity and 10 vulnerability of the system to minor variations in thermal image, the accuracy of the three dimensional model for re(Jistration of the face ima(~e, and the ma~nitude of sy~tematic and random errors.

Vl:. Selection of ~linutiae Since the face thermal surface can be distorted through changes in 15 expression, activities such as eatincg and talkino, tioht hats and other clothing~. sinus inflammation, and weioht g~ain and loss. the minutiae points to be extracted must remain fairly constant in spite of such chanoes or must be able to be filtered throu~h those changes. Section of minutiae as described herein provide minutiae well-suited to such factors.
Minutiae generator 104 selects minutiae from the facial thermo(~!am 200 after preprocessing as described above bv first positioning a circle of radius R on a thermal contour such that the contour intersects the circle, crossing it at two points and dividing it with equal area in each half ~e~t this circle is moved along the 2~
contour for as far as the contour can continue to intersect the circle at e.Yactly two points while maintainin(J an equal area on either side. If. in so movin(r the contour~
a location is found where further movement would cause tlle contour to intersectthe circle at onlv one point. the contour has ended. and the end point is desi(Jnated as minutia point. This situation tvpica]lv occurs onl~ at the edae of a facial ima(Je anc! only rarelv within the area of the face. If a location is found where further movement would cause the contour to intersect the circle at three or more points, the:re is an inflection point within the circle. It can be located by considering the slope of the contour within the circle relative to the face a~es. The point of 10 ma,~imum chancJe in slope is then designated as a minutia point. If a location is found where further movement would cause the contour to intersect the circle at no points. there is a small island area within the circle. The centroid ofthat island is designated a minutia point.
Referrin(J now to Fj(J ~;, there is sho~;n a facial image on which minutia I j points. e.g, 301. have been identified on facial thermo~;ram 200, as described above.

VII . Assi~gnment of Characteristics to ~ linutiae Once minutia points are selected. minutiae oenerator 104 assigns to each such point a label containin_ (~. y, z, c~. R B T), where ~ and y are the horizontal 20 and vertical displacements of the point relative to the facial a~es, z is the thermal valu.e of the point, ~ is the an(~le subtended b~ a tangJent to the thermal contour at the minutia point. R is the radius as discussed above, B is the number or value of the -thermal band in which the point is located. and T is the threshold imposed (if anv'l as discussed herein. In some environments, not all of these characteristics will '79 be used and in such situations they need not be assi_ned. Ho~ve~er, in some applications. these characteristics may advantageously be used for matching.
It should be reco_~nized that numerous variations in the operation and stnLcture of a minutiae ~Jenerator could be used. For instance, minutiae that are simply centroids of areas of constant thermal values could be used. Dependin_ onthe resolution of the thermal ima_ing device 102 on the order of 300 thermal contours mav typically be _~enerated for a face, leading to 300 minutiae. This number of minutiae mav be sufficient for identification purposes, depending on the application and environment in which system ] 00 is used. In some applications. it 10 may be of interest to identify faces seen in crowds or faces turned at any an ~le.
Parl:icularly in those applications, a significant number of minutiae points should be extractable so that even a partial face can be used for identification.
As another possibility, only centroids located near the center of the face~ or in concave areas of the face less vulnerable to artifacts due to ed~ges. could be used.
15 In other applications. minutiae may be derived using centroids of images ~~here all thermal values less than a threshold are maintained. but those hi_her values arecollapsed into one band. Centroids may be added to the set as the threshold is reduced. In such instance. each minutia point is characterized usin~ at least the (x, y z. T) factors mentioned above. where T is the threshold. In a v ariation on this ~0 mimltiae ~eneration technique, inflection points produced from such thresholdin rath, r than the centroids mav be used.
Still another variation is to use starl and stop locations from run len_th encodin of facial thermogram 200 to provide start and stop locations for thermalcontours. Each stop/start location provides a minutia point which is characterized '~5 by tlle (X?y,z) values discussed above.

~0 An additional approach is to desi(~nate undefined locations lJenerated b~
compression and subsequent eYpansion ofthe ima~e. Specifically. facial thermo~ram ~00 is compressed using wavelet or fractal-based methods and then eYpanded a(~ain. Because such compression techniques are lossv in a deterministic S vva~v, a minutia set may be defined as the undefined locations resultino frorn a cornparison of the ori(~inal ima(~e with the compressed-and-e~panded ima(~e. This approach provides an additional advantage of compressing the data used for recogmtlon, The wide variety of techniques for oenerating minutiae described above ] 0 provides an added measure of security, as one attempting to mimic thermal facial minutiae mav be able to do so if one technique for generatino minutiae is used by minutiae generator 104~ but not if another is used. Thus without prior knowledgeof the particular technique bein~ employed by minutiae oenerator 104, system 100becomes even more difficult to comprise than it otherwise mioht have been.
As mentioned above, it may be desirable that ail thermal ima~es be scaled to 2 standard size prior to processin_. It also may be desirable. dependin~ on the thermal imaging system used. that all thermal imaoes first be normalized to a standard thermal profile before processing. In alternate embodiments~ intended for various applications and various environments these preprocessino steps may ~0 si(Jnificantly increase accurac~ in recoonition or may merely impose unnecessary pro, essing overhead. For e~ample~ if system 100 is used in connection v. ith anoutdoor automated teller machine, thermal normalization mav be needed to deal with seasonally wide variations in surface skin temperature.

Minutiae Matcher 106 As mentioned above, minutiae generator 104 and minutiae data generator 105 are used to produce minutiae data sionals for a population of known persons.The data corresponding to these signals are stored in minutiae database 1 12.
5 Thermal imaging device 102 then obtains a thermal image of an unknown person 101 and minutiae generator ] 04 produces signals representative of the minutiae and minutiae data generator 105 generates data for the minutia for that person.
Once these signals have been produced, minutiae matcher 106 compares the signalsrepresentative of person 101 to signals from minutiae database 102 corresponding10 to minutiae data of known persons. In a preferred embodiment, minutiae matcher 106 performs three basic functions to obtain a match: alignment of the unknown face. comparison of minutiae data, and se9~tion of a match. Each of these functions 1~~
is described in greater detail below.

I. Alignment of Unknown Face Because there mav not be control over the position of the face of person 101 with respect to the field of view of thermal imaging device 102 when image is obtained, the orientation of the face may not be such that the facial axes are aligned to be horizontal and vertical. Thus, minutiae matcher 106 corrects the orientation by rotating the image such that the facial axes are horizontal and vertical. Next, 20 com/entional processing using a three dimensional model is applied to correct for any rotation or twist of the head. In a preferred embodiment, such processing ~ models the head as a sphere with a diameter equal to the apparent width of the face, and anti-distorts the image to provide a view which is normal to a surface plane across the forehead and upper lip and in which the center of the sphere coincides with the face center. In a conventional manner, the nose and chin are ignored so as not to disrupt positioning of this surface plane.

II. Comparison of Minutiae Comparison of the minutiae data of the unknown person 101 with minutiae data from known persons beg~ins by comparing locations of such minutiae. First, the locations of minutiae for a known face are considered, and denoted as M(K)i.Next an allowed positional error ~ is selected. as is determined to be appropriate for any given environment in which system 100 is used. The minutiae of the known face are then overlaid on the minutiae of the unknown face? denoted M(U)j.Any M(U)j that are not with ~ of one of the M(K)i are ignored. Any M(K)i w hich are not within ~ of on the M(U)j are i~nored. This leaves a residual set of minutiae pairs. If this set is empty, there is not a match between the two imag?~es.
Othlerwise, the characteristics of the corresponding points are compared.
1~ Depending on the application, any comparison technique that considers the characteristics (x, y, z, c., R, B, T) listed above may be used to generate a comparison metric. In a preferred embodiment~ only the positional differences are considered.
The simplest decision technique is to set a minimum number of pairs of ~0 corresponding minutiae for a potential match. If an unknown face and a known face exhibit a least the minimum number of corresponding minutiae pairs~ they are considered to be a potential match.
In an alternative embodiment, the !'\X and ~\y values for each pair of cor-responding minutiae are determined~ and the distribution of ~y with respect to ~x is then determined for the overall set of minutiae pairs. The standard deviation o:Fthat distribution is then compared against a threshold standard deviation to determine whether a potential match exists.
In still another technique. a new error measure ~' is introduced dependent not only on location but on thermal value (z). Minutiae pairs are only considered if they are within a certain thermal value difference '\z as well as have locations~ithin the distance error ~, thereby satisfying new error measure ~'.
Further levels of decision requirements can similarly be added to produce the desired level of confidence in the match for the application at hand. Each 10 possible comparison of the unknown face with known faces is performed~ and then the known images are rank-ordered according to the goodness of fit (e.g, closeness in metric) with the unknown face.

III. Selection of a Match Through experience with use of the s~stem on new images of known 15 persons, a threshold value is established to provide a desired ratio of false positive and false negative identifications appropriate to the particular application. In a preferred embodiment both self-correlations of multiple images of known persons and cross-correlations of different known persons in the database are used to help establish this threshold.
If onlv one known person meets the threshold requirement, that person is selected as the match. If no known person meet the threshold requirement, a failure to match signal is produced. If multiple known persons match the unknownperson to within the threshold difference, the best matching person is se!ected.Alternatively, if multiple images of the same known person are referenced in minutiae database 112. the person having the highest ratio of matches u ithin the top :number of best matches may be used. For instance if there are ten images ofeacl~ I;nown person in database 112, the top ten matching images determined by mimltiae comparison are considered. The person who is associated with the most of the top ten is selected to be the matching person. Additional levels of decision requirements may be added. either in a simple manner or iteratively, uith a determination after each level as to whether a match decision can yet be made.
In an alternative embodiment, minutiae matching is performed using tech:niques disclosed in U.S. patent application No. 07/984,514, filed December 2, 1992, and U.S. patent No. 0~/314,729, filed September 29, 1994. which is a continuation of U. S. patent application No. 07/9S4,514 both of which are herebyincorporated by reference in this application as if the entire contents of each had been fully reproduced herein In this alternative embodiment, flash correlation is used to match minutiae through a digitized artifact-producing technique. In this15 embodiment, the size of a minutia point is preferably increased to represent the possible error in its location, and minuti~e are replicated successively along the face axes to increase their density and thereby increase the discernability of the correlation artifact that indicates a match between tu o images being compared.
Such artifact is found to occur if any only if there is a match between two pixelized 20 images.
Other known matching techniques may alternatively be used in minutiae matc:her 106, with tolerances established for errors due to imperfect knov~ ledge of head position or distance, errors introduced by considering the head or face as a two dimensional surface or as a sphere, and other systematic and random residual25 errors. Some known fingerprint matching techniques may also be adapted to usewith minutiae matcher 106. By analooizin_ themlal contour to fingerprint ridges, the multiplicity of facial thermal contours may be treated in a manner similar to ma:ching many fingers per person. Alternativelv, specific areas of the face such as surrounding the canthi. may be selected and used alone for identification Depending on the resolution of the thermal imaging device ] 0~, several hundred 5 minutiae may be extracted from a facial thermal image. As noted above, lack ofprior knowledge of which facial features, and which specific matching techniquesare used for any particular application by svstem 100 increases the security of system 100 against being compromised by third parties.
~ For applications of svstem 100 to environments where legal proof of 10 identification is important. a classification scheme for faces may be useful. as fingerprints traditionallv have been classified into various classes for such applications. For e~;ample. whorls, arches, and loops are conventional descriptors applied to ridges in the center of a finger.
Another approach to classification of facial thermograms relates to obvious 15 characteristics for use in verbally descrihino a given facial thermogram. Such characteristics include whether the canthi are merged or separated; whether the thermal contour of the nose is relatively cold, hot, or normal; whether the nose is trapezoidal in shape or irregular in shape; the degree of thermal svmmetr of theforehead; and the degree of symmetry in location of thermal features in the mouth ~0 corners, the inner curves of the cheeks, the nose, the canthi, and the outer corners of the eyes. To be useful, such designations should remain consistent over variations in imaging equipment, environmental conditions, physiological variables, and other sources of errors. .~ccordingly, classification should not rely on features determined to be highly sensitive to such factors. Classifications based on overall ~ ~ image, e.g., those based on some ofthe distances between the features discussed in connection with Fig. 4 may be suitable for use.

Another approach is to use wavelet coefficients that produce the minimum difference between an interpolated wavelet-compressed image and the original irna(Je. Depending on how many classes are desired~ that number of wavelet ccefficient sets can be generated. Each image to be classified is compressed andthen restored using each of the sets. The image is assigned to the class correspondino to the set of uavelet coefficients which best restores the image to its original form.
Referring now to Fig. 6, classification of faces is achievable based on the nurnber of minutiae, their characteristics, and their distribution over the face. A
10 preferred classification method segments the face represented by thermouram 200 into a grid 401 of cells 402. Each cell is then classified based on the number of minutiae located therein. As an example, the facial thermogram 200 of Fig. 6 is divided into a grid 401. the cells 40'2 of which might be characterized as type A if the cell contains less than 3 minutiae, type B if the cell contains between 3 and 5 15 minutiae. and tvpe C if the grid contains more than 5 minutiae. A face can then be cla.ssified based on the number of cells of each type that are found. For instance, one classification scheme is based on the number oftype A and type C cells. If aface is divided in to 36 grid cells as illustrated in Fig. 6 classes could be desigrnated as nAmC, where n is the number of type A cells m is the number of type C cells, 20 n+m = 36 - p, and p is the number of B cells. Using this arrangement, 1260 classifications are possible. Alternatively, ranges of values can be considered to be wil:hin the same class.
As a further refinement to such classification, the degree of bilateral syrnmetry in distribution of type A cells and type C cells could be considered. If 25 the face is divided into four quadrants designated upper rioht, lower right, upper left, lower left, each quadrant having 9 cells, a metric for classification could lool;

at differences in the numbers of h~pe A and type C cells in horizontally or v ertically adjacent quadrants. Such metrics may be the absolute difference in minutiae between such quadrant pairs or may be simplified by merely indicating ~ hether alef't (or upper) quadrant has more, fewer, or equal minutiae as a corresponding 5 right (or lower) quadrant.
Other possible classifications are based on geometric values of, and ratios among, the points and lines described in connection with Fig. 4, once the face has been normalized as described above. In some applications. a combination of visual and thermal attributes may be employed for classification. For example, a ratio bel:ween the distance between left and right canthi centroids 201, 202 in facialthermogram and the distance between the left and right pupils as determined through visual imaging is found to be a useful metric for classification, as is the ratio between the distance from eye line 206 to horizontal central line 210 and the distance from a line connecting the eyes to the tip of the nose as determined by15 visual imaging, as is the ratio between the distance between left and right nostrils 20:3, 204 and the distance between the outer limits of the nostrils as determined by visual imaging.
The usefulness of facial thermal imaging in recognition applications is increased by appropriately encoding thermal facial ima_es so that consistent codes 20 are generated each time a facial thermogram of a person is obtained. Such a coding scheme reduces database search and minutiae matching overhead, thereby allowing faster processing using less expensive equipment. In a preferred embodiment, overlayin~ a grid on a face such that 144 cells cover the area of the face, and assigning a binary code to each cell, such t hat the cell is encoded with a 25 ~1" if the cell contains one or more minutiae and 0" if the cell does not contain any minutiae, is found in practice to yield _ood results. Since this encodin~J scheme preserves the relative location of each bit, it is strai~htforward to ignore selected bil:s in cases where oniy a portion of a face is imaoed due to obstruction. disguise, or orientation.
Use of such a -facecode'' also facilitates straightforward verification and 5 comparison techniques. In some verification applications, for example~ a requirement that 10% ofthe coded bits match may be considered sufficient to pn~vide a desired level of confidence. Simple difference comparison on a bit-by-bit basis, which is computationally extremely efficient, is sufficient to determine the number of correspondin(J bits between a code of an unknown face and that of a 10 known face. U~here multiple known faces exceed a threshold level of similarity, the one with the greater number of common bits is readily selected as a best match.
AlthoucJh the discussion above has been directed to thermal images of faces, it should be recognized that similar techniques and systems may readily be applied to images of other body parts in accordance ~ith the present invention. It 15 should also be recoonized that numerous other imagin(r modalities besides thermal imaging may be employed in accordance with the present invention, for example x-ray, NMR, MRI, and CAT scan imaginCJ. It should also be recognized that known schemes for pattern recoonition and (~raph matching may be applied readilyin accordance with the present invention, depending on the needs of a particular ~0 application.

Standardized Infrared Minutiae Co-ordinate Svstem (SIMCOS) The method and apparatus described above for facial minutiae extraction can be used to develop a standardized minutiae co-ordinate svstem for identification of medical patients and for diagnosis of medical conditions. Because an infrared camera operates at a distance from the patient and detects and records only radiant heat spontaneously emitted from the body surface, it constitutes a painless, non-invasive. passive method of recordinc~ patterns of bodv surface temperatures. These patterns have been found to depend upon the underlying S vascular structure and are unique for each person. Infrared identification therefore provides a method for uniquely identifying individuals under all lighting conditions, in,-luding total darkness. It is not prone to forgery or multiple identity deception and so provides convenient and highly secure identification of individuals. The ml. thod for generating repeatable registration points on the skin surface of the 10 hu.man body utilizes discrete minutiae points obtained from the thermal imag~es.
Visual characteristics of the body, such as size and shape and relative position of bcdy parts, are maintained in the infrared image. In addition, the details of the vascular s~stem are indicated by the distribution of temperature across the skinsurface. Current infrared cameras are sufficiently sensitive to temperature 15 variations that they clearly distinguish the skin directly overlaying blood vessels due to the thermal difference caused by the flow of ~-arm blood. The vascular structure appears as a white (hot) overlay of the circulatory structure on top of a grey scale image of the thermal map of the body. as shown in Fig. 7.
In Fig. 8, there is shown apparatus I OOa for processing infrared images to 20 yield repeatable minutiae points correspondin_ to specific vascular locations under the skin. The apparatus includes a thermal imaging device 102 for pr'oducing a thermal image I. A minutiae _enerator 104 and minutia data ,enerator 105 are part of the program memory 108 as is a minutiae overlay device 1 16. The set of minutiae obtained from any extended area ofthe body is unique to each individual.
25 In particular, facial minutiae are unique between identical t~ins. The same thermal minutiae are repeatedly e~tracted from a g~iven individual. They are overlaid by the overlay device ] 16 and annotated by an image processor 1 ~ 8 on the infrared image or on a visual or any image obtained from another medical sensor having the sameorientation to the subject. From the processor, the annotated image signal is de,livered to an output display/processing device wl1ich produces the thermal image 5 with overlay IO.
In Fig. 9, there is shown a modified apparatus 1 OOb to that of Fig. ~
wherein a reference minutiae identifier 120 is provided between the minutiae data generator 105 and the minutiae overlay device 116 to identify and specif reference points 151 in the thermal image and overlay IO. The reference points allow manual 10 or automated comparison, merging, or registration among a set of images talien at different times with different orientations or different medical instruments. Figs.
l Oa-c illustrate minutiae automatically extracted from a facial thermogram as the head turns. Current infrared cameras commonly produce 30 frames of video ou~ put per second and minutiae extraction and annotation can be performed in real 15 tin-le as those frames are generated.
Fig. I l illustrates alternate apparatus l OOc for generating a medical image MI of a portion of the body via a medical imaging device 177 such as an x-ray machine. The medical image can be annotated with a minutiae overlay to generate a medical image with minutiae overlay MIO from the output displavlprocessing ~0 device.
The inventive technique differs from visible recognition approaches in that it d.oes not merely sample a finite number of points on an image; it extracts points which have particular meaning. This provides increased resolution at the same time it reduces the degree of computation required. The essential features of the 75 technique are the uniqueness and invariance of thermograms~ the use of a passive imaging technique to obtain subsurface details, the use of automated minutiae 4]
e~;traction to match different images taken of the same individual. and the use of standardized minutiae locations to compare different persons or the same person as he grows from childhood to adulthood. The matching technique involves developing sets of corresponding points in two images. morphing one image into 5 the reference, and measuring the degree of morphing as an indicator of the amount of difference. Similar automated procedures are then used to verity that imager~ is from the same patient, identify a patient by comparison to a database of images,and compare images taken at different time and/or with different sensor modalities.
The minutiae extraction and annotation procedure locates the position of 10 eac:h minutia. In addition, it may note characteristics of each point such as a vector indicating the orientation of the corresponding blood vessel, a second vector indicating the relative orientation of the branching blood vessel, normalized apparent temperature measure. and apparent width of the corresponding blood vessels. As with some of the fingerprint minutiae matching machines, use of the 15 characteristic data can enhance the speed and accuracy of identification.
Fu]thermore, it can improve the accuracy and speed of automatic fusion of medical imagery.
This basic technique can be emploved on an area-by-area basis when poltions of the body cannot be seen or when significant changes have occurred in20 poltions of the thermogram such as when portions of the body have suffered external wounds. This would be done by seg-menting the thermo~ram to consider only the portions of the body in which the minutiae can be detected. Functionally.
this is equivalent to matching a latent partial fingerprint found at a crime scene to a full rolled print filed in the FBI system. The set of minutiae points, together with 25 characteristics which describe each such point and its relation to other minutiae is considered unique to the individual and persistent. for both contact fingerprints and 4~
thermal minutiae.
Verification that tw.o medical images are from the same person can be an end goal in itself or the first step in further processin_ the two ima_es to extract co-mparison data. Telemedicine applications, electronic filing systems~ insurance 5 claims processing, updating of medical records, and extraction of medical histories during emergency treatment are some of the situations in which it is essential to have a reliable, fool-proof method for positive identification of the subject and precise localization of ima_ed areas.
The use of infrared identification (IRlD) has several advantages over other 10 methods for recognition of persons. IRID operates reg~ardless of liyhting conditions. It requires onlv a single frame of ima_erv. taken in 1/30 of a second, for positive identification. and so can provide on-the-flv recognition during emergency admissions or evacuations. The imagery can be collected at a distance,~A~ithout causin_ a delav or inconvenience to the subject. No parts of the system I i come into contact with the subject. Since thermal images are essentially immune from variations caused bv illumination and shadows, it is not necessary to control those variables. IRID provides continuous identification and confirmation ve1ification of ID even in the dark. A cheaper, cruder form of thermal imaging can be obtained from use of heat-sensitive crystal sheets which require contact with20 the skin. More expensive, active imag~ing of the v ascular system can be obtained from laser doppler. Either of these could be the source imager for extraction ofminutiae; however, the operational advantages of passive thermal imacgers make them the preferred sensor.
Thermal minutiae can be obtained from commercially-available thermal 2i imaging devices sensitive in the 3 to 5 or ~ to 12 micron wavelength bands.
Images are this type are shown in Figs. 7. 10, 12. and 13.

Current infrared cameras produce a standard analoa or digJital output providing 30 frames per minute as shown in Figrs. 7-10. Tracking the minutiae from frame to frame assists in the exploitation of the dynamic IR imagery by allowing measurements to be made over time from the same body locations while 5 accommodating changes in position due to respiration, voluntary or involuntarymovements of the subject, and intentional or accidental variation in the position of thl im~ginC~ system. The use of infrared video imagery also allo~A~s the imagery to be recorded in real time for later analysis, and provides a self-documenting chain of custody identification of the person recorded, all u ithout the necessity for the ] 0 cooperation of the person being imaged.
Infrared imaging can be used to locate minutiae points over the entire body surface which correspond to intersection points and branch points of the underlaying blood vessels. This provides a built-in set of registration points on the body's surface, which can be annotated onto imaaes produced by any medical 15 sensor used in conjunction with the thermal imager. The registration points then ca-n be used to compare and combine medical images taken with different equipment at different times and under different conditions facilitating comparison of those images. Also the minutiae points provide reference points for continuous re-alignment of surgical instruments, radiation sources, and other diagnostic or20 treatment equipment. Since the infrared camera is totally passive, it can be used continuously during other medical procedures to overlay precise re(ristration points on the other images while also monitoring for overheating, shock hypothermia, renal failure, and other medical conditions. At the same time, the pattern of minutiae points superimposed on each image provides positive identification of the 25 patient. Such applications are of particular importance during telemedicine procedures.

The normal body is basically thermally bilaterallv symmetric. Side to side variations are typically less than 0.~5 degrees Celsius. This fact is used in assigning axes to the body's image. \A'here the sl;in surface is unbrol~en there is (~radual v ariation of temperatures across blood vessels, ~~ith the highest temperatures 5 across the body surface being directly on top of major b]ood vessels. Major thermal discontinuities occur at entrances to body cavities such as the eye sockets~
nostrils, or moth. These provide ~lobal reference points for automatic orientation of the thermal image. Local and relatively minor discontinuities in the sl~in surface occur at scars~ moles, burns, and areas of infection. The thermal surface can be10 distorted throu_h pressures and activities such as eatin=,~ exercising, wearin~ tight hats and other clothing sinus inflammation, infection~ wei(rht gain and ]oss, and body position. However, the minutiae points remain constant with respect to their position relative to the underlying blood vessels.
The technique for thermal minutiae extraction and matching can be 15 summarized as follows:

1. Current thermal image is digitized.

~. Current image is divided into pixels, ~~here the size of the pixel relates to the resolution or quality of the result desired.

3. Certain pixels are selected as minutiae points.

'~0 4. Each minutia is assigned a vector havin, magnitude and directional information in relation to the surrounding characteristics of the thermal image. Additional characteristics, such as type of 4~
minutia may also be recorded for each. Typically for each u;hole body thermal image, there uould be on the order of 1200 minutiae.

5. Set of minutiae vectors of the current image are compared by computer to the set of vectors of other images.

6. Comparison results are used to determine corresponding minutiae from the two images, and to morph or mathematically adjust one image with respect to the other to facilitate comparison.

7. Differences between the current imag,e and database images are computed for either the entire image or for areas of interest.

It is desirable that all thermal images in a database be normalized to a st;lndard thermal range and be scaled to a standard size during search and comparison procedures. Both normalization and scaling eliminate some minute arnount of identifying characteristics of a particular person or his condition.
However, the standardization procedures greatly aid in the exploitation of the database by reducing the need to calibrate every imaging sensor used to produce images which will be filed in, or compared to database images. For example, in ac:cident triage with no accurate ground truth reference in the scene ànd possibly use of inferior qualitv imagers, standardization to constant size and thermal range is appropriate in order to match against database. Furthermore, standardization facilitates use of simulated imagery for telemedicine and telesurgery applications.
For example, when incorporated into the military's automated battlefield medicalpod, real time normalized thermal minutiae can be used to properly position injections and application of external pressure to stop bleeding In addition. standardizing~ database imag~es facilitates comparison of imagery during c~ro-~th from childhood to adulthood, compilation of medical libraries of images from ]aro~e number of people~ and automated comparisons of current ima_er T against the vast libraries for diagnostic purposes. Standardization to a common use of thermal minutiae would also provide a common reference for comparing images obtained from different sensors which produce different resolution ima_es.
There is _reat utility to maintaining a summary medical record which could 10 be carried in encrypted form on a small token or card. It would include important medical history information, and would provide linkage into database holding more complete information. The use of thermal minutiae can be of assistance, since itprovide a standardized technique for segmenting the complete body~ using the thermal minutiae as nodes on a grid of finite elements.. The resultin_ cells in the 1~ gr.id would be coded based upon the compilation of all medical historv data relating to that area of the body, and compared against the standardized imagery and status of the correspondin-~ cell in reference model. If there were no entrv throughoutthe medical record for any imagery, diagnostics~ treatment. or injury involving that area of the body, or no sig~nificant deviation from the reference model, there would 20 be no data for the cell. If a known standard condition involving that area were known, then the standard code for that condition would be entered. Other codes would pertain to unknown conditions, continuing treatment, previous conditions su cessfully treated, etc. Conditions which are not localized, such as high blood pressure, wou]d by convention be assigned to specific cells within the body outline.
~ 2~ The identification techniques set forth herein can be used to diagnose and monitor treatment for burn victims, for stroke diagnosis, and in telesur_erv and telemedicine.
Blood vessels that car~ nutrients to the skin are destroyed ~vhen tissue is burned. A high pouered laser can be used to remove the burned skin, leaving, thehealthy sl;in intact. Laser li_ht is differentially absorbed and reflected by live and dead skin, such as in the area of a burn. Certain dyes such as indocyanine fluoresce when in contact with laser light and can be injected into a patient s blood to indicate healthy tissue. Alternately, an IR imager can be used to indicate dead skin vs. health~ skin, since the dead skin appears relatively cold as a result of having no functioning blood vessels. I\lore than 100.000 person per year in the US alone 10 suf'fer severe burns. The total cost of treating these patients exceeds $2 billion.
Treatment includes massive transfusions to replace the blood lost during, sur_ery to rernove dead skin. Blood loss is the main cause of death in burn patients.
Continued hea~,~,y bleeding often prevents a successful sl;in graft after the burned skin is removed.
Ideally performed. laser ablation kills a 100 micrometer-thin layer of skin belou~ the burned area. Due to the thinness, nutrients can still get through;
hou;ever. the layer stops the bleeding from preventing a good graft. IR minutiaecan be used to reposition the patient, monitor healing in specific areas, compare valious salves dressin_s. etc. used, and re-photo_raph the person, achieving~
20 standard imag~ing results reggardless of the skin tone of the patient.
Referring now to Fig. ] 4, there is shou~n an apparatus for maintaining the position of a surgical instrument 160 relative to a surg~ical site in a patient 162 during a surgical procedure. A thermal imaging device l 02 such as an infrared carnera, generates a thermal image output to a minutiae generator 104 to identif~,y 2 j minutiae in the vicinity of the surgical site. As described above with reference to Fig. 9, a reference minutia pattern is also _enerated for a statutory patient. A

sur7,ical instrument position detector 166 senses the position of the instrument and produces an output sio~nal corresponding therewith. The minutiae pattern for thepatient and the surgical instrument positions signals are delivered to a comparator 168. The relative position ofthe instrument to the surg~ical site is set as a r.eference 5 at time 0 at 170. Displacement ofthe patient and/or the instrument with respect to the reference at times other than O is detected by the comparator v.hich can produce an output 17~ used to reposition the instrument via a device 174 for proper orientation to the surgical site.
Each year, 500 00 American have strokes and 1507000 of them die, making stroke the third leading cause of death, and the major cause of disability amo~t ~~
adults. In the near future. new drug therapies may be able to return blood flow to stroke-damaged (ischemic) tissue protect it from permanent damag~e, and promote recovery of function. However, the primary care physician must be able to characterize the patient's acute neurological injury precisely enoug~h to guide ] 5 laboratory assessments and treatments. Stroke is a hetero_eneous ~roup of conditions with many causes, levels of severity, and clinical presentations.
Identifying the point of stroke damage and the extent of damage depends on a pattern of normal and abnormal findings. Computer tomographic (CT) findings often are normal during the first hours after ischemic stroke. Also, abnormalities found with neuroima_ing may be unrelated to the patient's acute problem. Further confusing the situation, seizures, tumor, and intercranial hernorrhage can mimic stroke. Treatment with anticoagulants or experimental clot-dissolving agents ~ould be contraindicated in patients with hemorrha_e but could be of value in some patients with ischemic stroke.
Continuous IR monitoring~ of the patient can possibly assist in detecting and tracking minute variations in blood flow patterns associated with the onset loc:ation, and severity of strol;e; and also associated with reaction to drugs and other treatments.
Within the next 10 vears, both military and civi]ian medicine are expected to make routine use of telesurgery where the patient and surgeon are not co-!ocated.
5 It is estimated that 90~/'o of the information a physician needs to know about a patient can be acquired and brought to him electronically. In laparoscopic surgery today, a surg~eon looks onl~ at video images without looking at the real organs at all. Dermatology and pathology are both already using electronic images as v. ell as x-ravs and medical records. Laparoscopic surgery is an electronic form of surgery.
10 Teleradiology, telepathology, and teleconsulation are already widely accepted electronic medical practices.
As telemedicine and telesurgery become more common, there will be more potential for error in identification of patients and the treatment to be performed~
and more need to document the precise medical history and treatment procedures 15 performed by a given doctor on a given day. Filing recall, and comparison of documentation collected over time by different sensors at different facilities ~ill need to be automated to a reater de ree, while yrotectin_ the privacv of the patients. The identification technique ofthe invention offers a low-cost re~peatable, non-invasive, passive system for standardization and registration of 20 many current forms of medical imagery while also offering an approach to highsecurity maintenance of files with immediate access in emergency situations.

Identification of Drug and Alcohol Usa(~e Many drug~s, including cocaine and alcohol, are vasoconstrictive substances which cause cooling of the skin surface. the resultant cooling is detected through passive imaging of the thermal energy emitted form the face. In Fig 15 is shown the thermal image of an individual who is substance free and in Fig. 16 is a thermal image of the same individual after the ingestion of alcohol.
The thermal imaging techniques of the present invention can be used to S delect substance use by individuals, even where the individuals's identity is unknown. This is accomplished by compiling databases of statistical analysis of thermal signatures obtained from clinical trials in which cooperating subjects have concurrent drug testin_ performed using urinalysis or blood testing along with thermal signatures obtained from known subpopulation but without concurrent testing by other means.
The vascular system supplying the human face typically exhibits thermal variations on the order of 7~C across the facial surface. Certain general features, suc h as hot patches in the sinus areas, relatively cool cheeks? and cold hair pertain to all facial thermograms. Other features such as specific thermal shapes in certain areas of the face are characteristic of a particular person Measured disturbances to other features, such as the general symmetry between two sides of a face, range of thermal variations in the forehead, peak temperature, size t~ canthi pattern, and variations in those disturbances over time, may be correlated with a high probability of drug or alcohol use.
Variations in temperature across the facial surface can be imaged by thermal cameras sensitive to wavelengths in the 3-5? 8-12, or 2-15 micron ranges.
Current cameras can provide thermal resolution better than 0.07~C and spatial resolution of better than .02 ", resulting in 65.000 to 265,000 discrete thermalmeasurements across the surface of the face. For most such cameras. that thermalmap is regenerated 30 times per second to produce either a standard video output~ which can then be recorded and processed on standard videotape equipment, or a direct digital signal which can be immediately input to a computer.
Certain dru_s appear to produce characteristics features in facial thermograms, which may be identifiable from detailed analysis of the structural patl:erns and distribution statistics. Furthermore, the rate of change at any point in 5 time may be a discriminator between chronic and recent use of each drug. Usingcun-ently available thermal imaging cameras, thermal signatures emitted from theface can be used to deduce changes in activity levels of specific arteries in the brain which are known to be affected by particular drugs.
The vascular system has a common structure in each person~ with known 10 pathways for instance from the heart to the brain, and known pathways betweenblood vessels in the face and those in the brain as shown in Figs. 2b and 2c. Using the SIMCOS technique, a set of standardized minutiae appearing in the face can be identified. Through clinical drug trials using known types, amounts, purity, andadministration techniques, the thermal effects over time at each such minutiae ] 5 location can be observed. The effect of varying the type~ amount, or purity of drug can also be observed. The effect on different people can be observed.
Since the thermal effects may be quite small and localized, it is important to utilize the SIMCOS method for identifying the precise minutiae locations in each subject.
That provides repeatability of measurements over time without requiring the 20 application of registration markers to the face, or the use of invasive techniques to repeatably find the same locations. Also, it provides a method for comparing corresponding locations in different subjects.
Statistical analysis of time-varying thermal signatures at each facial minutiae point before, during and after drug or alcohol administration provides a reference 25 dataset which represents the thermal effect of that substance under the protocol used A library of thermal minutiae substance effect signatures can be developed fo:r various drugs and other substances for wvhich screening is desired, includin~ for prescription and over the counter medications, tobacco and alcohol. Fig. 17 presents an illustration of thermal signatures associated with substance-free subjects. These may differentiate between se~, age, size, medical history, or other 5 characteristics of the substance-free subjects. In addition, thermal minutiae on nc,n-substance signatures ma! be developed for each person enrolled in to a system who will be subsequently scanned for substance use. At the time that the thermaldata is collected, urinalysis or blood testing can be performed to assure that the subject is substance-free.
Subsequently, each time a person enrolled in the system is scanned the same facial thermal minutiae are located and the corresponding thermal signatures computed. They are compared against that subject's own substance-free thermal si(~,natures. If there is sufficient match, no further analysis is required. Otherwise, the thermal si ~,natures are compared against the librar~ of substance effect 15 signatures. If there is sufficient match, the system provides an output to the system operator that a potential substance detection has occurred.
If desired? relevant data from the svstem may be transmitted to a medical revie~h officer for a final determination. Such data could include the current thermal imagery of the person's face7 reference imagery from the database taken of 20 the same person when he was known to be substance-free, the thermal si~natureca.lculations which lead to the systems's determination of a possible substance detection, the library reference thermal signature which was used by the system to make that determination, the system's calculated confidence in the determinationthresholds which were set by the system manager relative to the infrared camera 25 sensitivity and other system factors, medical data on file about the person including known or self-reported use of prescription or over the counter medication, or a past history of substance use. The ~ledical Review Officer determines whether toaccept or override the system's determination, or specify that further testing such as urinalysis is to be performed.
The technolo_y of this patent involves clinical studies in whicll known amounts 5 of controlled substances are administered to subjects whose thermal images aremonitored and stored to provide archival references. Analysis of those images is used to determine the time-varying thermal effect of specific drugs at specific minutiae sites on the face, which are specified by anatomical landmarks. After an individual ingests a drug, changes in his or her thermal signature gradually occur until a thermal "climax"
10 is reached after which the signature gradually returns to its normal state. In chronic drug users, permanent physiological changes may occur such that there is no lonsger a smooth total decay of the apparent drug-induced effects. By processing a significant number of thermal images, thermal signature markers are identified and related to standardized vascular system locations whose thermal variation are highly correlated 15 with use of the particular substances. References may be developed for an individual, for a class of individuals grouped by age or other characteristic, or for a general population.
A general determination of substance-free vs. substance-influenced classification may be based upon data collected on the thermal effects of various 20 substances of interest. In addition, certain substances produce characteristic results which may be identifiable from detailed analysis of the thermal signatures associated with facial minutiae, and/or with distribution statistics from those signatures.Furthermore, the rate of change at any point in time mav be a discriminator between chronic and recent use of each drug. Techniques for processing sequences of thermal . 25 images may enhance the visibility of bilateral asymmetries, ano.~malous static ~ conditions, and unusual time-varying trends in the thermal signatures associated with .

specific minutiae locations in the face to indicate activity levels of specific arteries in the brain which are knou;n to be affected by particular druos. Therefore, when asubstance is known to affect particular functions. vascular pathwavs to the corresponding brain areas should be analyzed for related thermal signatures at 5 minutiae points along the pathways, under the assumption that increased activity at the brain site will be found to correlate with increased v asomotor activity along pathways leading to that site, as evidenced by thermal changes.
In order to best compare images from different people and under different conditions, facial thermograms must be standardized and registered to common 10 coordinates The preferred approach is to use the standardized infrared minutiae coordinate system (SIMCOS) technique which locates standard minutiae points on each facial thermogram. In its preferred embodiment, the SIMCOS minutiae correspond to anastomoses which are connections or branchings of major superficial veins and arteries in the facial area. I~pproximately ] 75 such points exist in the face.
15 A subset of the SIMOS minutiae which relate to blood vessel or areas of the face affected by a particular substances is selected. The substance activist at the brain site will be found to correlate with increased vasomotor activity along pathwavs leading to the site, as evidenced by thermal changes.
The collection of differences Ibetween the time-varying thermal signatures for 20 the substance-active minutiae, compared to the collection of time-varying thermal signatures for the same points in the absence of the substance, represents the marker for that substance. For each substance of interest, a marker may be developed for a particular individual, for a class of persons grouped according to some criterion, or for a more general population. The substance-free marker can likewise be developed 2~ for a particular individual class of persons, or general population.

In subsequent screening of a known individual for a particular substance. his current therrnal image(s) are analyzed to extract substance-active minutiae which can be seen in the available ima_e(s). I'he set of thermal signatures is compared to the substance marl;er collection and substance-free marl;er collections. Measures of5 similarity are calculated for individual, class, or general population comparisons for each substance of interest. ~arious correlation associated with substance use, or with substance-free references in the system librarv. Normalized temperatures vs. time waveforms for each minutiae can be compared with the corresponding waveform fromthe reference. The w aveforms are s,lid along the time axis until the best fit is found, 10 since it is not known when, if at all, the person being screened may have used that substance. Another correlation approach involves sampling the thermal waveforms and producing a matrix of values, where one dimension of the matrix is the number of minutiae used, and the other is the number of temperature samples over time. The reference library can include wider matrices, involving longer time period than is I j practical for an operational screenino system. The comparison between the collected matrix and the reference matrix would use a digital shifting and difference calculation to find the best area of match.
A measure of goodness oi' match is made between the collected thermal signatures and the si_natures for each substance under each protocol in the library.
20 The system manager selects a threshold to be applied to each comparison, such that matches which are closer than that threshold will cause the system to issue a notice of possible substance detection.
The results of comparison with the different markers may be recorded or stored or output to decision marker ;. Alternativeh~, thresholds may be automatically 2j applied to the calculated differences to render a pass/fail or cleanlunder-influence determination. The statistical estimate of confidence in the determination can also be presented.
The apparatus for dru_ and alcohol detection is shown in Fi_. 18. Three primary functions are performed with the apparatus: Enrollment. Reference Si_natures Development and Screenin_. System components may in ~ eneral be rendered as 5 software, hardv.are. or firmware elements.
Prior to automated operation of the Identification and Detection (ID&D) system, a human operator termedl a System Manager must perform set-up and initialization of the system, which he does via the System Manager Interface, which includes a monitor, i~eyboard, and possibly a printer and other peripherals such as a 10 mouse which are normal]y associated with personal computers. A System Mana_ermust confirrn the identity of the enrollee 4, and input the associated identification information into the person identifier and condition identifier database 6 within the system processor 8. The enroliee's current and past medical historv data is also input to the system. including the results ol~urinalysis or blood tests to detect substance use, I 5 current use of illicit substances, and other information which may bear upon alcohol and drug testing results.
The enrollee stands or sits at designated location facing the infrared camera 12and within reach of the event trigger 14. ~'hen the event trigger is engaged by either the enrollee or System Manager, the output from the infrared camera is sampled by 20 the frame grabber 16 and the resulting frame stored in its buffer. At the same time, the camera output is recorded on a video cassette recorder 18 which incorporatesannotation of the date. time, location, and identifv of the enrollee.
The image in the frame grabber and buffer is processed by the face locator 20 which determines that the image includes a sing~le face which is in focus and of a 25 suitable size and position. If the image is not suitable according to software criteria established witl1in the face locator, a new image frame is grabbed and the process repeated until a suitable ima(~e is obtained.
The imaoe is then processed by the minutiae extractor and identifier 2?~11iCh locates the SIMCOS minutiae points and extracts their positions on the image and the corresponding apparent temperatures. Additional frames are grabbed and processed5 for a period of time selected by the system manaoer.
The extracted minutiae locations and corresponding temperatures are processed by the thermal signature extractor 24 which generates for each minutiae point the thermal variation over time. For enrollment purposes, a single frame may suffice. However, multiple frames over a period of seconds should be taken in order 10 to help calibrate and factor-out noise in the system.
Personal data about the enrollee is transferred to the enrollment database 26, along with the thermal signatures e~tracted for the enrollee. If the enrollee is l;nown to be substance-free, the thermal si<~natures are also transferred to the substance-free signature database 28 stored by enrollee and also to the substance-free thermal 15 signature database stored by classification of enrollee 30.
If the enrollee is known or found to be substance-influenced, the thermal signatures are instead transferred to the substance-related thermal signature database stored by substance and also to l:he substance-related thermal signature database stored by protocol 34. The definiticns of protocols will relate to clinical trial used for ?0 developing reference thermal signatures? and may also include self-reporting classifications such as heavy regular user of cocaine", or infr~quent user of marijuana but not within the past rnonth."
The extracted thermal sionatures of the enrollee are also transferred to the thermal signature comparator and statistical analyzer 36 which compares the 2~ signatures of the enrollee ~ith other signatures in the databases. If the enrollee's signatures vary too much from the others in the same substance-free class or from others in the same substance-related or protocol-related databases, then the svstem manao~er may request revieu by the medical revieu officer 38 prior to including the enrollee's data in the database.
If no anomaly is detected in the enrollee's thermal signature then the enrollee 5 is instructed as to hou- to activate the system for future access and screening. If a personal identification number is to be used, that PIN will be assigned. If voice recognition, ID card, or other technique for identification is to be used, thoseprocedures will be taught. The enrollee is now enrolled in the system.
~ The substance use identification and detection svstem reguires databases of 10 thermal signatures from substance-free and substance-related images. During clinical trials in which substances are administered under ri(Jorous protocols, the same apparatus may be used to _enerate the reference databases.
A clinical trials investigator will serve as system manager . He will perform set-up and initialization of the system, the s~ stem mana_er interface ~ u hich includes 15 a monitor or other display~ l;eyboard and possibly a printer and other peripherals such as a mouse which are normally associated uith personal computers The system manager must confirm the identit:y of the enrollee 2, and input the associated identification information into the person identifier and condition identifier database 6 within the s~stem processor 8. Since the same person may be enrolled several times 20 under various protocols~ he is termed the subject'' to emphasize that he may have several separate files within the enrollment database ~6. The subject's current and past medical history data is input to the svstem the first time he is imaged. During each di~renl test involving different substances and/or protocols, the specifics of the protocol used, as well as recent medical data including the results of urinalysis or 25 blood tests to detect substance use, will be included in the person identifier and condition ID Buffer 6 The contents of that buffer are transmitted and stored with the . CA 02229630 1998-03-16 results of the thermal signature extractor ~4.
The extracted thermal signatures of the subject are transferred to tlle thermal signature comparator and statistical analyzer 36 uhich compares the signatures of the subject with other signatures in the databases, if any. If the subject's signatures vary 5 too much from the others in the same substance-related or protocol-related databases, then the system manager may request review by the medical revieu; officer 38 prior to including the subject!s data in the databases.
The thermal signature comparator and statistical analyzer 36 processes the thermal signature to establish composite signatures or common characteristics which ]O represent each ofthe databases: substance-free signatures for each person enrolled in the system; and substance-free signatures for each designated class of enrollees, such as: people under age 16. people 17-~0, people ~1-30, people over 30, cigarette smokers, non-smokers, people takino, heart medication, diabetics on insulin, athletes.
vegetarians, social drinl~ers. etc. If a sufficient number of subjects is used, likewise 15 composite signatures or common characteristics will be calculated for each substance in the database of substance-related signatures. and for each protocol.
After the end of clinical trials, frames are extracted from the archi~al videotapes and used in place of imagi es directly talien by the infrared camera. Each extracted frame is processed by the svstem, and the thermal signature comparator and ~0 statistical analyzer 36 selects the best matching composite signature and classifies the frame accordingly. First, the thermal si~nature collection of that frame is classified as better matching the composite for all substance-free trials or the composite for all substance-related trials. Second, if the subject in the frame is judged to be substance-free, the best matching class of database 30 is then selected. If the subject is judged ~ ~5 to be substance-related, then the besl matching substance of database 3~ is selected.
Third, in the substance-free case, the system identifies the best-matching enrolled person's signatures within the class selected. In the substance-related case, the system identifies the best- matching protocol for the substance selected. Since the true classification of each subject on the video tape is known the performance of thesystem i~an automated mode can thus be determined, and the error rates associated ~, with the classification can be calculated. Improvements to the system can then be made according to standard techniques for statistical analysis, including the use of neural nets to adjust weightings consideration of the data from certain minutiae more or less than others, normalizing the -thermal variations or not, and normalizing time intervals based upon the size or metabolism of each subject.
When those system parameters have been adjusted to maximize the correct classification from the video tape archives, the resulting reference signatures are stored in database 38 and used to speed the screening of future enrollees duringoperational use of the system.
An enrolled person seeking entry through the system as an entrant stands or 1~ sits at designated location, facing the il~rared camera 12 and within reach of the event trigger 14. When he is in position, he activates the event tri_g~er. The output from the infrared camera is automatically sampled by the frame grabber 16 and the resulting frame stored in its buffer. At the same time, the camera output is recorded on a video cassette recorder 18 which incorporates annotation of the date, time, location, and 20 identity of the enrollee.
The image in the frame grabber and buffer is processed by the face location 20 which determines that the image includes a single face which is in focus and of a suitable size and position. If the image is not suitable according to software criteria established within the face locator"a new image frame is grabbed and the process~ 2~ repeated until a suitable image is obtained.
The image is then processed by the minutiae extractor and identifier 22 which locates the SIMCOS minutiae points and extracts their positions on the image and the corresponding apparent temperatures, Additional frames are grabbed and processedfor a period of time set into the system by the system manager, The extracted minutiae locations and corresponding temperaIures are 5 processed by the thermal signature extractor ~4 w hich generates for each minutiae point the thermal variation o~ er time.
The entrants's ID is transferred to the thermal signature comparator and statistical analyzer 36. along with the thermal si~Snatures extracted for that entrant, The thermal si_nature comparator and statistical analyzer compares the signatures of 10 the entrant with database of reference signatures 3 S. The closest reference signature is selected and the entrant is designated to ha~e the same classification, That classification is transmitted to the sy~,tem manager interface. which may grant access through manual intervention or through automatic control of an access portal.
Depending upon the particular cla,sification selected~ or the amount of ~ariation 1 j between the entrant and the selected reference signatures, the result classification may be sent either automatically or through manual inter~ention to a medical review officer 40 for a final determination. The entrant's file from the enrollment database ~6 is also sent to the ~RO to provide background information.
Statistical anal~sis of drug or alcohol use ~ ithin a random population is also ~0 encompassed by the present invention, This is performed by scanning a crowd and locating faces therein for analysis.
Various standard methods fi~r locating faces in an image frame can be used.
A particularly useful approach uses an ellipse detector to find relatively warm ellipses (thermal faces) within a relatively cool background. The ellipse is located within the ~5 expected hei~Tht range for humans, the detected temperature is within the e~pected range for human faces. and charact:eristics common to all facial thermograms (hot canthi regions, s~mmetr~ of the e~,es nostrils, ears cheel;s, etc.) are not violated.
Each detected face in each frame.is evaluated to determine if it meets the quality requirements for further processing. Requirements include the facial image being in focus, being large enou_h to pro~ide adequate resolution of the facial minutiae, being 5 oriented close enough to full face forward and beino free enough of blockages including beards eyeglasses. and intervening obstructions~ such that a sufficient number of facial minutiae can be extracted from the facial image. The specific requirements are dependent upon what substances are to be detected, in ho-~ small a dose, and after what period of time. Faces w hich do not meet the qualitv requirement 10 are not further considered. Those which are qualified are assigned unique tao,s.
The next frame is then processed and qualit~ faces are detected as above.
Each such face is then compared to the faces in the previous frame, or to those fi~s 1 from the previous frame which are close enough in location that they could be a particular face in the current frame l\,Iatching is performed using the facial minutiae 15 matching method. If a current and prior face are determined to be the same, then they are given the same tag. This process continues with subsequent frames being likewise analyzed. In general. a particular tagged face will move across and then out of the field of view. U'hen the face is no longer seen by the camera, then the thermal signatures associated ~ -ith each of its minutiae in each of the frames are combined and 20 matched against a reference datab,~se for non-substance and substance-related signatures.
The system is designed for stand-alone operation. IT is deplovable for programmable periods of time. during which it will anal~ze and classify each face which appears within its field of ~ie~. The svstem will not routinely record or store 25 the thermal images. although provisions are made to do that during testing and evaluation of the system in order to allow for improvements to be made in the system ;md compnrcd with CalliCJ Ics~ s. The Ollt]lllt t~om the systeln will bc ~raphical reslllts such as shown in Fi,g. 19. The cum~ative detection inde~; of thc y-L~isrepresents the numbcr of pcrsons whu Ihe syslern t~.ctimates have used marijuana, cocaine, or heroin in an amoun~ l wi~hin a ~i,ner~ e. which results in ~ residual S ~evel in~ir~tcd by thc .~;-axis value ~I Lbe lime of Ihe an~iysis. Tbe x-a~c~s represcnts he confidence le~el of dru;, signatulc ;IId;CaUOn~ which i.~ rel~ted to the detection l~r~rislon ofthe testing and ~nslysis procedules. Separale cur~es indicate the specific dnl~c detected, and ~ cornpositc survcy indi~ dle~ delection of any ot'the subst~ces.
Dl-~ to the ~requent u~ of combinations of drugs~ the composite cllrve Is expected to 10 he mc~re sisnificant than its components.
The system is tested usir,lg ~;nownpop~ nls of tirug users, an~i its results colllp~ul to ~Irinalysis results. Thc co..~l A~son is us~l lo selec~ thresholds fnr cystem decisions on d~csific,atiorl of therm31 si~;nnturcs. ~bc s~slc~ an be deployed within a high in~ensity ~n~g l~afIiclil~s s~e~, and its rcsults co~ red lo o~her current 1~ csuma~es of drug usa~t~. In that area.
The apparatll~ for sr~tisti~ql analysis is shown in .l~ig. 20. Three pnmary 1;-nclionc are pelfo"l,ed with the apparatus: face r~uicition and tagginB; f~ nalysis ~md ~ ;sific~rion; and stati~ti~l population analysis.
All i~dred camera 42 i~ r~ositioned such that persons in t}lc populatioa l~J b~
20 scanned ,e,cller~lly enter, uanCYe. se, and e~t the c3mer~'s ficld of ~icw 44. Thc infral-ed ~nera produces a ~ lence of frames usin~ eilher dircct digital outpul ~lr ~;ame-~betl Yideo output 4t-, wbich is stored in the frDmc buf~er stora~,e 48.
]'roccssing of lhc imagery is inib'at~.n ~y an jniti~ r mecl ~nisrn 50 which may bc a proximity delcc~or~ mo~ion de~ect(lr, nr other sensor used to detcct thc possiblc prcscnce of humall~ within Ihe field of ~,iew The system 1~ ssor 52 is compri~efl nf live components. The facc locator 54 applies rules to the buffer stored image to identify all faces in the frame. The face quality check 22 applies additional rules to deterrnine if a given face provides sufficient information in terms of focus, resolution, position, and number of minutiae available.
Each qualified face is tagged and then compared with qualified faces in the preceding S frame? using the SIMCOS technique, to determine which faces have already been seen and tagged.
Faces which are seen in a given frame, may have been temporarily blocked in the previous frames, and so any face ~hic~ is not blocked. turned, or otherwise unqualified part ofthe time. Comparison against earlier qualified faceslcontinued for 10 a period oftime which is considered a reasonable maximum time for transversing the filed of view. The face tracker 54 reassigns tags so that the same face receives the ~Q~
frame tag in subsequent frames. There are two reasons to be concerned about P~
consistently applying the same tag to each face. First, drug and alcohol detection accuracy improves when a face is analyzed for a longer period of time. Second, 15 statistical accuracy improves when each person is counted only one time, regardless of their position or speed, or their being within a group.
Once a face has exited the i;eld of view, the thermal signature extraction processor 60 produces time varying apparent temperature signatures for each facial minutia seen over the course of time of all frames in which that face was seen. The 20 composite thermal signature for that face is produced at 62.
The analyzer 64 of the composite thermal signature compares each imaged person's composite thermal signature to a database of signatures associated withknown substances and protocols or with known subpopulations such as cocaine addicts or alcoholics. More particularly, a signature from a thermal signature database 25 from deployment 66 is compared with a signature from the reference thermal signature 68 in a comparator 70. the output of which is delivered to a thermal signature classifier 72. The analyzer output is delivered to a statistical analyzer 74 forprocessing the thermal signatures by batch to produce an output summary to the output report generator 76 for the output device 78. A statistical sufficiency analyzer 80 evaluates the report and determines whether the analysis of the population up to that point is statistically sufficient for the intended purpose. If not~ additional frames are analyzed. If the analysis is sufficient, then the system is re-initialized by the initializer 50 starting a new collection of faces and leading to a new statistical evaluation.
~f ~AD
If the population throughe~*t is well-known, the statistical analyzer 74 can be ' q 10 programmed to run for set period oftime before the batch is considered complete and control is passed to the output repolt generator 76. Other criteria for determining a batch size can be used. including counting the number of tags assigned or the number of detected suspected drug users or the number of detected substance-free persons.
While in accordance with the provisions of the patent statute the preferred ] S forrns and embodiments ofthe invention have been illustrated and described, it will be apparent to those of ordinary skill iII the art that various changes and modifications may be made without deviating from the inventive concepts set forth above.

Claims (33)

WHAT IS CLAIMED IS:

1. A system for recognizing individuals comprising (a) an imaging device for producing a first signal representative of sensed characteristics of the individual;

(b) a minutiae generator connected with said imaging device for receiving said first signal and producing in response thereto a second signal representative of minutiae of the individual, the minutiae corresponding to specific branch points of blood vessels of the individual;

(c) a minutiae database for storing minutiae data characteristics of minutiae for each of a plurality of known individuals and for producing a third signal representative of minutiae of each of the plurality of known individuals; and (d) a minutiae matcher connected with said minutiae generator and said minutiae database for receiving said second and third signals and for producing a fourth signal when a match occurs between the minutiae for the individual and for one of the plurality of known individuals,

2. A system as defined in claim 1, wherein the sensed characteristics are non-visual characteristics,

3. A system as defined in claim 2, wherein the sensed characteristics are thermal characteristics.

4. A system as defined in claim 3, wherein said imaging device comprises an infrared camera.

5. A system as defined in claim 3, wherein said minutiae are limited to those onthe face of the individual.

6. A method for recognizing an individual, comprising the steps of (a) sensing non-visual characteristics of known individuals;

(b) identifying minutiae of each of the known individuals in response to said sensed non-visual characteristics thereof, said minutiae corresponding to specific branch points of blood vessels of the known individuals;

(c) sensing non-visual characteristics of a new individual:

(d) identifying minutiae of the new individual in response to said sensed non-visual characteristics thereof;

(e) determining a distance metric for each of the known individuals with respect to the new individuals in accordance with one of the identified minutiae and the data characteristics of the identified minutiae thereof; and (f) determining a match between the individual and one of the known individuals in accordance with the distance metrics.

7. A method as defined in claim 6, wherein said non-visual characteristics are thermal characteristics.

8. A system as defined in claim 7, wherein said minutiae are limited to those inthe face of the individual.

9. A method for classifying an individual, comprising the steps of (a) sensing characteristics of the individual;

(b) producing a normalized representation of the individual in response to the sensed characteristics thereof;

(c) identifying minutiae of the individual in response to the normalized representation. said minutiae corresponding to specific branch points of blood vessels of the individual;

(d) determining a correspondence between a grid of cells and the normalized representation; and (e) classifying the individual in response to co-location of at least one of said minutiae and at least one of said cells, thereby identifying the individual as corresponding to a class

10. A method as defined in claim 9, wherein said classifying step includes encoding the identity of the individual using a plurality of bits each corresponding to one of said cells, setting one of the plurality of bits to a first state in response to the presence in a corresponding cell of one of said minutiae and to a second state in response to the absence in the corresponding cell of any of said minutiae.

11 A non-invasive method for identifying medical patients, comprising the steps of (a) identifying minutiae on the body of a known patient, said minutiae corresponding with branch points of blood vessels;

(b) storing a collection of minutiae data characteristics of the minutiae for the known patient in a memory to define a reference collection, said reference collection being unique to the known patient;

(c) sampling the minutiae data of an unknown patient in a selected location of the unknown patient's body to define a sample collection;

(d) comparing said sample collection with said reference collection, whereby the identity of the unknown patient can be confirmed when said sample and reference collections correspond.

12. A method as defined in claim 11, and further comprising the steps of (e) detecting any changes in said sample pattern with respect to said reference collection; and (f) analyzing the changes to diagnose the occurrence of a medical event in the patient.

13. A non-invasive method for diagnosing a medical condition in an individual, comprising the steps of (a) establishing a reference collection of series of time-varying minutiae data corresponding with a plurality of known medical conditions. respectively, said minutiae data being derived from characteristics of minutiae associated with branch points of blood vessels in a human being;

(b) generating time-varying minutiae data from the individual having an unknown medical condition; and (c) correlating said generated minutiae data with said reference minutiae data to diagnose the medical condition when a match between said generated and reference data is obtained.

14. Apparatus for identifying, the use of a substance such as alcohol and drugs by an individual, comprising (a) means for identifying minutiae data in the face of the individual which are responsive to the ingestion of a substance by the individual, said minutiae data being derived from characteristics of minutiae associated with branch points of blood vessels of the individual;

(b) minutiae data generator means for generating a first collection of minutiae data of a substance-free individual and a plurality of second collections of minutiae data of an individual after the ingestion of a plurality of different known substances;

(c) means for storing said first and second collections of minutiae data in a database; and (d) means for comparing a subsequent collection of minutiae data of an individual with said first and second collections to determine whether the individual is substance free and to determine what substance the individual has ingested where the individual is not substance-free.

15. Apparatus as defined in claim 14. wherein said minutiae data identifying means comprise ( I ) an imaging device for producing a first signal representative of sensed characteristics of the individual; and (2) a minutiae generator connected with said imaging device for receiving said first signal and for producing in response thereto a second signal representative of minutiae of the individual, said second signal being delivered to said minutiae data generator means.

16. Apparatus as defined in claim 15, wherein said imaging device comprises an infrared camera for scanning an individual over a period of time, said camera responding to thermal changes in said minutiae resulting from the ingestion of asubstance. whereby said first and second collections of minutiae data comprise thermal signatures of substance-free and substance-affected individuals, respectively.

17 Apparatus as defined in claim 16, and further comprising processor means connected with said minutiae data generating means for generating time-varying thermal signatures from the scanning of the individual by said infrared camera.

18. Apparatus as defined in claim 17, wherein said processor includes correlation means for determining the closest match between a subsequent collection of minutiae data and one of said first and second collections of minutiae data.

19. A method for recognizing whether an individual has ingested a substance suchas drugs and alcohol, comprising the steps of (a) identifying minutiae data in the face of the individual which are responsive to the ingestion of a substance by the individual, said minutiae data being derived from characteristics of minutiae associated with branch points of blood vessels of the individual;

(b) generating a first collection of minutiae data of a substance-free individual and a plurality of second collections of minutiae of an individual after ingestion of a plurality of different known substances;

(c) generating a subsequent collection of minutiae data of the individual; and (d) comparing said subsequent collection with said first collection to determine whether the individual is substance-free.

20. A method as defined in claim 19, and further comprising the steps of comparing said subsequent collection of minutiae data with said plurality of second collections of minutiae data in the absence of a match with said first collection of minutiae data in order to identify the substance that the individual has ingested.

21. A method as defined in claim 20, and further comprising the step of correlating said subsequent collection of minutiae data with said first and second collections of minutiae data in order to determine the closest match therebetween.

22. Apparatus for determining what portion of a random population has ingested a substance such as drugs and alcohol, comprising (a) means for generating and storing reference collections of minutiae data corresponding with a substance-free individual and with substance-affected individuals, said minutiae data being derived from characteristics of minutiae associated with branch points of blood vessels on an individual;

(b) means for generating current collections of minutiae data from individuals in the random population; and (c) means for comparing said current collections of minutiae data with said reference collections of minutiae data in order to determine the portion of the random population that is substance-free and the portion of the random population that is substance-affected.

23. Apparatus as defined in claim 22, wherein said reference collection generating means generates different collections of minutiae data in accordance with different known ingested substances, whereby the portion of the random population which has ingested each known substance can be identified.

24. Apparatus as defined in claim 23, wherein said current collection of minutiae data generating means comprises ( I ) an infrared camera for scanning and tracking individuals of the random population, said camera producing a first signal representative of a thermal signature for each individual, and (2) a minutiae data generator connected with said camera for receiving said first signals and for producing a collection of minutiae data corresponding therewith, respectively, whereby a plurality of collections are generated, one for each individual in the random population.

25 A method for determining what portion of a random population has ingested a substance such as drugs and alcohol, comprising the steps of (a) generating a plurality reference collections of minutiae data corresponding, with substance-free and substance-affected individuals;

(b) generating current collections of minutiae data for individuals in the random population, respectively; and (c) comparing said current collections of minutiae data with said reference collections of minutiae data in order to determine the portion of the random population that is substance-free and the portion of the random population that is substance-affected.

26. A method as defined in claim 25, wherein said reference collection of minutiae data include different patterns corresponding with known ingested substances.
whereby the portion of the random population which has ingested each known substance can be identified.

27. A method for annotating an image of the human body. comprising the steps of (a) generating an image of the human body with an infrared camera; and (b) superimposing on said image the location of minutiae detected by said camera, said minutiae corresponding to specific branch points of blood vessels of the body.

28. A method as defined in claim 27, and further comprising the step of identifying particular minutiae to serve as reference points.

29. A method as defined in claim 27 and further comprising the steps of generating a medical image of the human body and annotating the medical image with said minutiae.

30. A method as defined in claim 29, and further comprising the step of identifying particular minutiae to serve as reference points.

31. Apparatus for annotating an image of the human body, comprising (a) means for generating an infrared image of the body;

(b) means for generating a pattern of minutiae from said infrared image, said minutiae corresponding with specific branch points of blood vessels of the body;

(c) means for generating a medical image of the body; and (d) means for annotating said pattern of minutiae on to said medical image.

32. A method for maintaining the position of a surgical instrument relative to asurgical site during a surgical procedure, comprising the steps of (a) identifying minutiae in the vicinity of a surgical site on a patient, said minutiae corresponding with specific branch points of blood vessels of the patient;

(b) generating a reference minutiae pattern for a stationary patient;

(c) generating a reference position of a surgical instrument with respect to said reference pattern;

(d) detecting deviations from said reference pattern owing to movement of the patient and from said instrument reference position owing to displacement of the instrument, and (e) repositioning the instrument with respect to said reference pattern to accurately position the instrument at the surgical site.

33 . Apparatus for maintaining the position of a surgical instrument relative to a surgical site during a surgical procedure, comprising (a) means for identifying minutiae in the vicinity of a surgical site on a patient, said minutiae corresponding with specific branch points of blood vessels of the patient;

(b) means for generating a reference minutiae pattern for a stationary patient;

(c) means for generating a reference position of a surgical instrument with respect to said reference pattern.

(d) means for detecting deviations from said reference pattern owing to movement of the patient and from said instrument reference position owing to displacement of the instrument;
and (e) means for repositioning the instrument with respect to said reference pattern to accurately position the instrument at the surgical site.