WO1989002639A1

WO1989002639A1 - Process for automatic classification and search of fingerprints

Info

Publication number: WO1989002639A1
Application number: PCT/BR1988/000012
Authority: WO
Inventors: Arysio Nunes Santos
Original assignee: Arysio Nunes Santos
Priority date: 1987-09-15
Filing date: 1988-09-06
Publication date: 1989-03-23
Also published as: EP0333788A1; JPH02501196A; AU2300488A; BR8705132A

Abstract

Process for automatic classification and search of fingerprints, allowing the automatic and inexpensive storage and retrieval of fingerprint sets into a computer memory for the purpose of Personal Identification in both Criminal and Civilian usage. The process is compatible with the usual Henry or Vucetich Systems, and utilizes opto-electronic or visual methods of acquisition and encoding of the macroscopic features of the fingerprint patterns. It takes into account the statistical errors resulting from print distortions, misprints, skin abrasions and shrinkage, among others that may occur.

Description

TITLE OF INVENTION; PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS DESCRIPTION

Among the methods utilized for Personal Identification, that of fingerprint impressions (dactyloscopy) is by far the most popular, given its simplicity, ease of processing, and the fact that they are absolutely unique from person to person. Fingerprints are also virtually indestructible, and yield a permanent register which is easy to file and classify for use in personal identification,or that of criminals.

The process of obtaining fingerprint records consists of daubing the ten fingers with black ink, and applying them to paper. The pattern of ridges is thus imprinted on the paper which is of standardized form and quality. These regis- ters thus obtained are called decadactylar fingerprint records, and form an individual register when the individual's name, date of birth, affiliation, date of registration and register number are added to the form, accompanied or not of a photo¬ graphy and other pertinent data. The field of application of the decadactylar register is in both Criminal Identification and in Civil Identification. Criminal Identification is effected by the comparison of the latent fingerprints often left in the scene of the crime by the criminal with the fingerprints on file. Civil Identifi- cation is used for the purpose of Individual Identification, and is effected by taking the decadactylar prints of the person in question and comparing these prints with those on record.

The practical problem with dactyloscopy is that the files often contain tens of millions of individual fingerprint sets, which must be scrutinized individually by experts cr by machines, in order to sort out the correct one; so that the process becomes extremely tedious and time-consuming. The present invention eliminates all these problems in a thorough way. It allows not only an easy classification and instant sorting of an individual' s fingerprints by automated means, but it also allows a permanent updating of the records, as well as remote operation by means of low-cost telematization and computerization of the data.

There are two methods presently employed in dactylos- copic identification. One is that of Henry/Vucetich, and the other is the so-called. AFIS (Automatic Fingerprint Identifi¬ cation System) . The Henry/Vucetich System is currently utiliz¬ ed in many countries, having been introduced in 1894 in Argen¬ tina by Vucetich and in 1900 in London by Henry. It is based on the fact that individual fingerprints form several basic patterns such as Arch, Loop, Whorl, Double Loop and Accidental. These are, to an extent, independent for the 10 fingers, so that we have a possible 5 = 9,765,625 combinations.

However some of the patterns are extremely rare, whereas others such as loops are most abundant. Moreover, the patterns for a given individual are not wholy independent, so that some combinations are extremely abundant, whereas others are rare. Thus the type E3333/I2222 (all ten fingers consisting of Ulnar Loops) occurs in something like 10% of the cases, whereas its mirror image (I2222/E3333) almost never occurs in practice.

In large cities, files range by the tens of millions of records, that of the FBI exceeding 100 million individual decadactylar records. So, the most common combination such as E3333/I2222 can have from 1 to 10 million records, and individual culling becomes impossible. The solution attempted was to adopt sub-divisions of the individual patterns in what is called the Extended Henry (or Vucetich) System. Accordingly, Arches are sub-divided into Tented Arches and Plain Arches; Whorls are sub-divided into Left-Convergent and Right-Conver¬ gent, or Oval and Circular; Loops sub-divided into Ulnar and Radial (Internal and External) and/or Small/ Medium/Large, and so on. Besides these subdivisions, a sub-classification iε also done based on the so called minutiae, the fine details such as ridge branchings and terminations.

The problem with sub-divisions is that classification becomes somewhat uncertain and subjective, for the sub-divisions merge smoothly among each other. Hence, many intermediate types are often classified in one sub-type and searched in another, depending on the operator and on the quality of the print. In a large file such as that of the FBI, some 40,000 searches and new classifications are performed daily by hundreds of operators, so that the work has to be performed in a rush; and quality and finesse are lost, while many clas¬ sifications are missed or wrongly placed.

Worse still, most of the searches occur for the most abundant types, whose number on file is also extremely large and sub-division necessarily finer and subtler, as well as more erratic and subject to arbitrariness. A thorough job requires that all records in a given sub-division be searched, along with all those in adjacent sub-divisions where the record searched many have been misplaced by chance. Everytime a search has to be made, the operator has to go back to a full cycle of hypotheses of possible erroneous sub-classifi¬ cations, and search all of these thousands of records on each of them. Besides, the numerical encoding of an individual's fingerprints becomes impossible, and tele atic processes can¬ not be utilized in practice, except by image transmission via Telefax or Telephoto.

This desperating situation found a partial solution in the AFIS machines mentioned above. These are sophisticate image acquisition and processing devices based on super¬ computers that are fast enough and large enough to process and store the hundreds of millions of images of the finger¬ prints and pick out and compare the hundreds of individual minutiae on each of them.

The problem of the AFIS is their huge cost and com¬ plexity of operation and maintenance. They are able to work with a single fingerprint — instead of the decadactylar systerr. of Henry/Vucetich — and are thus adequate for latent finger¬ print identification in criminal investigation. But they are not adequate for Civil Identification (decadactylar) because their speed cf operation is too slow and the quality of the prints it requires far exceeds that usually available on the files. Attempts to use AFIS systems in Civil Identification have failed miserably, particularly in underdeveloped countries, where quality control and maintenance are far below the re- quired standards.

The present invention of a Process- for the Automatic Classification and Search of Fingerprints is specifically intended for Civil Identification purposes, given that it is based on a novel process for encoding the decadactylar finger- print pattern of an individual that is essentially unique and imune to error. It is compatible with the Henry/Vucetich primary classification, while dispensing with the erratic sub- classifications. Its use results in an encoded alpha-numerical pattern that can easily be remotely transmitted or registered, dispening with the large memories and the complex algorithms required in the image processing and storage utilized with the Standard AFIS.

The present invention is also highly tolerant to poor quality of the prints and ofthe classification process, since the encoding process utilizes a high degree of redundancy in the data. By redundancy we mean the use of superfluous pieces of data that can be ignored without loss of selectivity. The processing and storage of the data is simple, fast and compact, so that data aquisition can be done in field work with consultation of the central file via radio or telephone, and in situ classification of the suspects by standard techni¬ ques of dactyloscopy. Alternatively, when performed at the base, the classification process is automated via light-pens and bar-code readers, adapted for the use of encoding finger- print images projected on an episcope and which are entered directly in the computer's memory.

Once encoded, the fingerprint record can be dispensed with, and reserved only for the (rare) cases where certainty isrequired, as, for instance, in criminal investigations or post-mortem identifications. Since the present process is compatible with the Henry/Vucetich systems in current use, its implantation does not upset the usual routines or files, which can proceed normally even if the system that implements our process fails. The alpha-numerical code that results from the utili¬ zation of the present process can be used as a personal attribute for identification or individualization purposes, since it is absolutely unique, intransferrable, unadulterable and directly verifiable by standard dactyloscopic techniques that can be applied even remotely. This application is extre¬ mely useful in airport terminals, customs and police stations, and even personal identification systems that are able to automatically acquire and process data or transmit them to a central for consultation. The resulting alphanumerical code can be further encoded via bar-codes or magnetic dot systems that render falsification essentially impossible for lack of compatibility.

The great problem with the Henry/VBucetich extended systems is obtaining sub-divisions that is not subjected to self-correlation among an individual's prints, so that the sub-division is really effective in separating different in¬ dividuals. After a detailed study of the most diversified sub-classification and encoding possibilities for fingerprints, as well as of their statistical distribution in practice over large numbers of persons, we finally developed the process of automatic alpha-numerical encoding which is the object of the present invention.

Our process — which offers all of the above-mentioned advantages and suffers of none of the disadvantages of previous processes and systems — essentially consists in the opto¬ electronic aquisition by means of a mouse, a light-pen or a bar-code encoder or other such opto-electronic means of the individual configuration of parameters such as the ridges that mediate between specific portions of the fingerprints. This is done for each of the ten fingers of an individual, and the results — together of those of the Henry/Vucetich Basic System and/or its sub-divisions, are utilized to form the alpha-numerical code that individualizes the record in question. The resulting code is stored in a computer's memory, whence it is retrieved and compared with the queries by means of a specially developed algorithm that utilizes a novel searching method capable of dealing with statistical errors of encoding. Instead of the minutiae searched in the standard AFIS, our process utilizes the macroscopic features of the finger¬ print pattern. It thus uses an average result which essentially eliminates statistical errors. Among the parameters that our process utilizes for encoding — which can be obtained either visually or by means of an opto-eletronic device coupled to a computer — we utilize the average value of the ellipticity of the whorl patterns; the radius of curvature of the central line of the loops or of the limiting line (ridge) of the loop or of the apical line of the arches; the strike of the medial line of the patterns; the total area of the central system of lines; the individual types of delta (such as black or white or dotted); the number of ridges between (say) the center of the pattern and a major feature such as the delta or the inter-phallangeal fold, and so on.

As an example of an encoding procedure by means of the decadactylar process we invented, consider the following implementation. The individual fingerprint is opto-electroni- cally entered in a computer, and the desired parameter is traced out by means of a mouse or a special-purpose light- pen. The value of the parameter — f.i., the radius of curvatu¬ re of the limiting line of a whorl or loop or the number of lines between the delta and the center of the print — is thus automatically acquired and computed, and is entered intc the computer's memory as a two-digit number.

The procedure is carried on for the 10 fingers, so that a group of 10 numbers oftwo digits are obtained. The record is also classified (visually) by the standard Henry/ Vucetich primary system, hich is entered as a Primary Code, whereas the above-mentioned ten parameters form the Secondary

Code. A typical result would thus read A1234/V4321 - 33 43 17

28 19 46 78 67 56 02. The first ten characters are the Henry

Code and the last ten sets of two digits each, the Secondary

Code obtained by the above procedure.

7 For a large file of say, 1G records, we have some-

5 . . . . thing like 10 records of a given Henry/Vucetich type, wmcn are treated as a group in the computer's RAM. This is easily seen to be of moderate size, even if we add other data such as name, registration number, etc., either directly or in a collateral storage device. The Secondary Code yields (100) =

10 possible combinations, so that the a priori probability

5 of a random coincidence in the 10 records on a group is onl ,y ₁1_n05/_/11_Λ0100 = 1-0«-95 However, the parameters' values are not all wholly independent or exempt from statistical errors of measurement, so that a more realistic value of the effective range is some¬ thing like 10 units for each finger, yielding 10 effective combinations and an actual probability of 10 /10 = 10 for a random coincidence. Thus, only once in every 10 queries would an additional record show up in the search. But this is no problem, for the operator can either verify the actual prints visually or check further data on record in case this rare event ever occurs in practice. Such ancillary data can be weight, height, sex, race, age, etc..

A specific problem ofthe method we invented and herein describe is that of the statistical errors in a parameter's determination by either visual or opto-eletronic (automatic) methods. An individual's fingerprints are subject to abrasions, erosions, distortions, burns, wrinkles and shrinkage that tend to alter the value of any characteristic parameter from take to take. Even though such fluctuations do not prevent the individualization of an individual's set of print, as demonstrated above, they require that the encoding and search- ing algorithm be able to cope with statistically varying numbers. In other words, the desired register must be found, despite the fact that the entry register differs from it due to statistical error.

This is again a novel and crucial feature of the present invention and process of encoding and searching deca¬ dactylar fingerprint sets. In order to obtain the desired tolerance to deviations between entry and register we treatec the parameter's value as a random number, and quest forthe register that, within a given Henry/ Vucetich sub-set, yields the highest probability of being correct. If other candidates are acceptable, they are also displayed, along with its a priori probability, in decreasing order.

The algorithm we developed for one cf the possible- implementations of the present invention is as fellows. The computer program converts all characters of the code into binary numbers and stores them in ordered fashion. The first register is A1111/A1111 00 00 00 00 00 00 00 00 00 00 and the last one V4444/V4444 30 30 30 30 30 30 30 30 30 30, where we supposed that the characteristic parameter's value ranges from 0 to 30 (2⁵bits) .

Each group of a given Henry/ Vucetich type is treated as a block, and is brought into the RAM as a single group of

5 typically 10 registers. Since only the primary code is utilized, this group is essentially imune to error. In case of doubtful, intermediate types, they are classified in both of the possible alternatives, so that they are always encoun¬ tered. If desired, accessory data (sex, race, weight, age) are also entered to reduce group size or accuracy. Next, the Secondary Code is searched as follows. Each of the two digit numbers of the code is classified as either Small, Intermediate or Big — respectively S, I, B — and is entered in the respective sub-division. The width of the In¬ termediate region encompasses the range of expected error, say ± 1 digit, so that 5 ranges from 0 to 15 while I goes from 14 to 16 and B from 16 to 30. The groups increase from SSSSSSSSSS to BBBBBBBBBB, to a total of 2¹⁰ = 1024 groups. The computer program picks each of the two digit numbers of the Secondary Code of the quested register and determines if it is S or I or B. If it is either S or B, the program selects the corresponding half of the memory; and if it is I, it picks both. Thus, all parameters lying in the fuzzy interme¬ diate region comprised by the window I is picked out, since it is in either of the two sub-divisions which are, both, brought to the RAM.

For instance, if the set in question is, say, SSSISSSS SI, the following blocks will be brought to the RAM of the computer: SSSSSSSSSS, SSSBSSSSSS, SSSSSSSSSB, SSSBSSSSSB. This binary method of statistical quest is theoretically the quickest and the most economical of memory. But several other variants are possible with only small detriment. And only rarely will more than a few groups be fetched into the RAM. For instance , if there are 5 fingers with I values of the encoded parameter, we get 2 = 32 groups. Since the groups are evenly divided— for the value of I is taken at the median of the distribution — for a file of 10 5 registers (secondary code) we have 105 ÷ 103 = 102 registers per group. With 32 groups as a probable maximum, we get something like 3200 registers (maximum) in the RAM, a moderate a mount.

Next, the registers brought to the RAM are individually compared with those of the quested one. The difference .Δi between the i. th two-digit parameter of the quested print and that on the memory is computed, and so is the product of all (1 + Δi) factors which represents the inverse of the proba¬ bility of a deviation of Λi in each finger. The candidates are listed in decreasing order of a prioriprobability, to a desired level of statistical confidence.

Claims

1 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" for use in Personal Identification of both Civilian and Criminal purposes, characterized by the use of an alphanumerical code obtained from the ten fingerprint sets of an individual, by visual or optoelectronic methods, and consisting of a Secondary Code obtained from the ordered grouping of parameters such as the ten individual numbers of ridges between center and delta or between that between the intraphalangeal fold and the center of the pattern, the Se¬ condary Code being compled to the Henry/ Vucetich Primary Code, as well as to other accessory data such as Henry/Vucetich extensions, or sex, age, weight, height and so on.

2 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" according tc Claim 1, and characterized by the fact that the Secondary Code utilized in the encoding of the fingerprint decadactylar set utilizes such characteristic parameters as the ellipticity of the whorls, the inclination and radius of curvature of the axial line of the loops or whorls or arches, the height-to-base ratios of thearches and any other such averaged parameter of the ten-fingerprint sets.

3 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" according to Claims 1 or 2, and characterized by the fact that the algorithm of search ofthe code registered in the memory of a digital computer consists of a statistical method coupled to a searching window that encompasses eventual statistical errors in the determination of the characteristic parameters of the fingerprints, so that all registers are recovered even in the presence of substantial errors of encoc- ing, down to a prefixed level of priori probability which is also directlv calculable. 4 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" according to Claims 1 or 2 or 3 characterized by the fact the filed register is divided into two portions in order to enhance speed; the encoded fingerprint being quested in a sub-file that only contents the registration number and the registration number, which not being subject to statistical error, is subsequently utilized to search the other section of the memory that contents the other portions of the data on record. 5 - "PROCESS FOR AUTOMATI ^"CLASSIFICATION AND SEARCH

OF FINGERPRINTS" according to Claims 1 or 2 or 3 or 4, chara¬ cterized by the fact that the Alphanumeric Code obtained in the encoding of the fingerprints is further codified by bar¬ codes, magnetic dots orother such encrypted methods, so that they can only be read by means of special devices, and are thus rendered immune to forgery wherever they are employed, as in the verifiable encoding and documenting of an individual's identity.

6 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" according to Claims 1 or 2 or 3 or 4 or 5 and characterized by the fact that the encoding for both the classification and the search of the fingerprint sets is done automatically by the computer, via optoelectronic methods of image acquisition and of pattern recognition or of iterative computer graphics, helped or not by the operator, that traces on the screen display of the fingerprint, the desired line or pattern with the help of a special light-pen or mouse or bar-code encoder while the computer automatically determines and registers the desired parameter such as number of ridges, ellipticity, radii of curvature, and so on.

7 - "PROCESS FOR AUTOMATIC CLASSIFICATION AND SEARCH OF FINGERPRINTS" according to Claims 1 or 2 or 3 or 4 or 5 or 6, and characterized by the fact that the computer program for the classification and search of fingerprints utilizes an interactive menu and internal instructions that orient the operator in the search of the optimum parameters to utilize with a given pattern and memory capacity.