WO2002080088A1

WO2002080088A1 - Method for biometric identification

Info

Publication number: WO2002080088A1
Application number: PCT/CN2001/001452
Authority: WO
Inventors: Jiesheng Wang
Original assignee: Wang, Qin
Priority date: 2000-09-25
Filing date: 2001-09-25
Publication date: 2002-10-10
Also published as: CN1346116A; CN1144158C

Abstract

The present invention comprises the following steps: 1) image separation: detecting the edge, outline or region of the analyzing object from the somatic tissue image; 2) regional standard transformation: establishing standard mesh over the image region that extracted; 3) carrying out regular transformation on the texture in the standard mesh; 4) encoding with equilong and maximal entropy; coding the regular features of the texture on each subregion; 5) feature code analyzing: analyzing two sets of codes obtained by processing two different images of the same human body biological tissue using above two steps: if the two sets of codes are similar, the two images are identical, or else are different. Said mehtod is suitable for recognizing different kinds of somatic biological tissue to obtain available images via image acquisiting equipment, and the recognizing error rate is far below prior art because of analyzing images in microscopic view.

Description

Human biometric identification method

The invention relates to a method for identifying human biological characteristics. Background technique

Human biometrics refer to human physiological tissues that can be used to identify individuals, such as fingerprints, retinas, iris, face, face shape, DNA, etc. Compared with non-human biometric authentication such as passwords, passwords, seals, etc., human biometrics are unique, unchanging, and inseparable basis for identity authentication. Human biometric authentication is natural and reasonable. Identity authentication method. In human biometric recognition technology, digital fingerprint recognition systems have a long history of development. In recent years, other digital recognition systems have appeared one after another. Looking at the existing human biometric recognition technology, we can see the following characteristics:

Independence

A feature of the existing human biometric recognition technology is that there are different recognition methods for different physiological tissues. The technologies and products developed according to these methods naturally constitute different independent systems. There are independent image acquisition equipment, independent algorithms, independent software and chips, and even separate application areas and use environments.

2. Use pattern matching method

Another feature of the prior art is that pattern matching methods are mostly used. These methods are either digital methods developed based on manual comparison methods or methods formed based on pattern texture image processing technology. Because the methods used are empirical or cannot analyze images on a microscopic scale, they have poor accuracy and reliability. In addition, the codes generated by this method are usually unequal-length codes, which are not suitable for matching, which increases the recognition difficulty and reduces the recognition accuracy. Invention Disclosure

Aiming at the problems existing in the prior art, the present invention will provide a new identification method. This method takes a specific human physiological tissue as an identification object, acquires an image using a collection device, and authenticates the identity through a special processing of the image. This method is suitable for the identification of various human biological tissues that can be used to collect available images through image acquisition equipment. It provides general recognition methods and techniques for different human biological tissues, and analyzes the images on a microscopic scale so that The rate of misunderstanding is far lower than the existing methods and technologies.

The task of the present invention is achieved by the following technical solutions:

A human biological feature recognition method using an image acquisition device and a computer, the steps are − (1) Image separation

Use an image acquisition device to collect images of human biological tissues, and use a computer to detect the edges, contours, or areas of specific analysis objects from the given images of human biological tissues in order to separate and extract the analysis objects;

(2) Regional standardized transformation

Use a computer to establish a standard grid on the separated image area, that is, the analysis area, and perform the area normalization transformation;

(3) Regularization of texture features

Regularize the texture in the standard grid with a computer;

(4) Equal-length, maximum entropy coding

Code the regularized representation of the texture on each small area with a computer and use it as a digital representation of the image texture on the entire analysis area;

(5) Feature code analysis

Use a computer to analyze two different images of the same human biological tissue using the methods in (1)-(4) above to determine whether the two images come from the same human biological tissue of the same person. Identity verification, using a computer to analyze two different images of the same human biological tissue using the methods in (1)-(4) above to determine whether the two images are from the same human biological tissue of the same person That is, identity authentication is performed. If the corresponding code points with different values in the two groups of codes occupy less than or equal to the specified value, the two images are the same, otherwise they are different. Brief description of the drawings

The method of the present invention will be specifically described below with reference to the drawings.

Figure 1 is an iris image of a human body;

Figure 2 is a fingerprint image of a human body;

Figure 3 is a facial image of a human body. The best way to implement the invention

The invention is a human biological feature recognition method using an image acquisition device and a computer. Because to recognize specific human biological tissues and their images, the image of this part of biological tissues must first be separated from the background. Then, texture features are extracted from the separated image, and these features are analyzed and compared with the archive data in the archive (database). Finally, the identity authentication conclusion is made for the subject holding the physiological tissue. The present invention proposes a new implementation method for each of the sub-processes in the above general process, thereby becoming a brand new recognition technology. These new methods and technologies are shown in the following table − Purpose content method

Image Separation Edge / Area Detection Custom Boundary / Area Detector

Spatial analysis

Feature analysis time domain analysis regularization of texture features

Tabular mechanism

Feature representation

Feature recognition Feature recognition Feature code analyzer A detailed description of the method steps of the present invention is as follows.-

(1) Custom boundary / area detector

A. Problems to be solved

Detect edges, contour lines or regions of a specific analysis object from a given image to separate and extract analysis objects.

B. Method

Obtain images of specific human biological tissues through image acquisition equipment. For convenience of description, this two-dimensional grayscale image of biological tissue is represented by f (x), where X represents the position of a pixel point in the image, and f (x) is the grayscale of the point. We call the automatic boundary / area detection mechanism in the recognition process a custom boundary / area detector, which can be implemented by software or chip. The working principle of the custom boundary / area detector is described as follows:

a. Customize the structure of the boundary or area to be detected: If the boundary to be detected is composed of straight line segments, a linear boundary detector can be defined. In general, the detector is not necessarily linear, and it is usually a line integral (for detecting the boundary) or area score (for detecting the area) of the image f (x) with λ as the parameter, which is denoted as F (x, ), Where Fj (x, A) is a function of f ( _x ). Among them, λ is a geometric parameter related to the shape of the boundary or region to be detected, such as an angle, a radius, a radius of curvature, and the like.

b. Smoothing the detector: To illustrate the concept of detector smoothing, we first describe the smoothing of image points.

Smoothing of image points: Each point (current point) in the image should not be viewed in isolation, and the effects of surrounding points on it should be calculated, that is, smoothing should be performed. The point that needs to be considered for the current point is called the current point's action point, and the area formed by the action point is called the scope. If only linear effects are considered, the scope can be a line segment with the current point as the midpoint, or IJ, a rectangle or other shape with the current point as the center. The magnitude of the influence of each action point on the current point can be different and needs to be given by a smoothing function. If it is considered that the influence of nearby points is large and the influence of distant points is small, a Gaussian function G (x, σ) is generally selected as the smoothing function, where σ is a scale factor. Detector smoothing: determine the scope of the detector F (x,), and use G (x, σ) for each point (detection point, current point) on F (x, A) Smooth over the domain.

c. Derivate λ or calculate the change: The detector takes λ as a parameter. The physical meaning of using λ as a parameter is that there may be multiple groups of pixels in the image that can constitute the boundary shape defined by the detector (a group of points is called a possible boundary), but generally only one of them is true boundary. Different possible boundaries have different lambda values. In order to search for the true boundary from the possible boundaries, it is necessary to calculate the change in the overall characteristics of each possible boundary (for example, the total gray value of the possible boundary) when λ changes, that is, the derivative of each possible boundary with human variables , Or calculate the amount of change.

d. Calculate the extreme value of the derivative or change: From the nature of the image, the true boundary has the property that its change reaches the maximum among all possible boundaries. Therefore; the true boundary can be determined by calculating the extreme value of the derivative or the amount of change.

(2) Regional standardized transformation mechanism

A. Problems to be solved

Even for the same physiological tissue of the same person, images of different sizes may be obtained at different times and in different environments, and other properties of the images may be different. In addition, the original shape of the image may cause many inconveniences to the analysis process. In order to process different images of the same physiological tissue and make the analysis area have a shape suitable for processing, the analysis area must be standardized for transformation. We still use f (x) to represent the image of a specific physiological tissue. The transformation requirements can be summarized as follows (may only require meeting some of them):

a. Transform an image of an analysis area on a plane to another area;

b. transform the boundary of an analysis area from one shape to another in the image; c. standardize the texture features inside the analysis object;

d. Transform the image from one coordinate space to another coordinate space.

B. Method

In general, not all transformations in the above requirements are required. For the item d in the above requirements, after the coordinate system before and after the transformation is given, it can be implemented by ordinary coordinate transformation, which will not be discussed here. The following general methods are given for the above items &, b, and c:

a. Establish a standard grid on the analysis area according to the characteristics of the actual problem; to establish a grid, two principles must be followed: one is to establish the same grid for the same physiological tissue, and the other is to divide the grid according to the geometry of the area. In the grid. If the geometry of the analysis area is already suitable for analysis, other transformations can be omitted after the standard grid is created.

b. If you want to transform the geometry of the analysis area, you must transform the line segments in the grid one by one to the target area. In this case, the original boundary contour (set to C1) and another contour line to be converted (set to C2) are known. The transformation is performed in the following manner: Let the intersection point of the line segment a and the boundary contour line C1 being transformed be xl, x2, and a (x) be on the line segment a , The value of the transformed image f (X) in the target region is determined by two known functions hi (x, xl), h2 (x, x2), hi (x, xl), h2 (x, x2 ) Is called a normalized processing function.

(3) Regularized transformation mechanism of texture features

A. Problems to be solved

There are two different types of textures. One type is pattern texture. In this type of texture structure, there are texture primitives, and the texture primitives or local structures composed of primitives are roughly uniformly repeated in a larger range. The other is textures with no regular structure. Similar textures have the characteristics of disorder, non-characteristics (no obvious texture primitives), non-uniformity (non-uniform repeating arrangement), and uncertainty (the same type of physiological or physical organization has an uncertain texture structure). The texture of human biological tissues can all be regarded as unstructured texture. For such textures, it is very difficult to extract geometric features. The present invention provides a new method for regularizing such texture features, which can uniformly process different human biological features and tissue textures, which makes the present invention completely different from the existing technologies such as fingerprint recognition technology, and can different biological features Recognition is unified.

B. Method 1

Let's consider the analysis area after the area feature transformation as R (x, y), Ri is the ith small area divided by the transformed grid, and (x0, y0) is the geometric center of this small area . A coordinate system is established on each small area with (x0, y0) as the center, (xm, yn) is the coordinates of any point in Ri, and _g (xm, yn) is the gray level of point (xm, yn). We will perform a regular transformation on the texture on each small area in this coordinate system. The mechanism of regular transformation of texture features is as follows.-A. Smoothing the image: In the implementation method of "Custom Boundary / Area Detector", we have smoothed the image to detect the boundary or area. Here, we smooth the image again to regularize the texture features. The method is as follows: The gray value g (xm, yn) of each point is weighted (that is, multiplied by a factor, or weight coefficient). In addition to the coordinates xm and yn, these weight coefficients can also be selected from a sine function, a cosine function, an exponential function, a hyperbolic function, or other functions.

b. For a given point, find the smoothed sum of g (xm, yn) multiple times with different weight functions, that is, the multiple weighted summation, generally an even number is used.

c. Find the weighted sum ci described in a., b. above for all points in Ri, then ci is the regularized characterization of the texture features of that point on region Ri.

C. Method 2

Regularized table method: The weighted smoothing algorithm for the regularization of texture features can be summarized as a set of numerical operations to obtain the numerical coefficients for performing spectrum analysis on the divided small area and compile them into a table. Selecting numerical coefficients in a region and performing simple arithmetic operations can realize the regularized characterization of the texture features of human biological tissues in this small region. This method is called a regularized table method. Regularized table method is only a simplification of method 1 There is no substantial difference in the calculation method.

(4) Equal-length, maximum entropy coding

A. Problems to be solved

The regularized characterization of the texture features on each small area is coded to be used as a digital characterization of the image texture on the entire analysis area, and as the basis for feature recognition. Because the analysis area has been standardized, the code is equal length.

B. Method

a. The weighted functions (sine function, cosine function, etc.) in the regularized transformation of texture features can be mapped to each other (in special cases, it can be positive and negative), and the regularized representation of each other Treated the same.

b. In order to meet the above requirements, the invariant f in the mapping must be taken as the encoding threshold: the weighted value ci f of each point on the region Ri, the corresponding code point value is 0, otherwise the value is 1.

c. Regarding the determined human biological tissue, in the regular transformation of texture features B. The smoothing processing of the same number and the same weight function is used, and the same-length code is obtained. It can be proved that this coding method is the maximum entropy coding and contains the maximum amount of information.

(5) Feature code analysis

A. Problems to be solved

Given two different images of the same human biological tissue, two sets of codes (characteristic codes) are obtained after processing using the methods in steps (1) to (4) above, and the two sets of codes must be analyzed to determine the two Whether the images come from the same biological tissue, that is, for identification.

B. Method

a. Calculate the Hamming distance hd of the two sets of codes, that is, the percentage of the corresponding code points with different values in the two sets of codes among all the code points;

b. Determine the authentication threshold δ according to statistical characteristics. When hd S, they are considered the same, otherwise they are considered different.

The following are specific examples of identifying human biological tissues using the method of the present invention: 1. Iris recognition

The iris image is shown in Figure 1: In the picture, the ring-shaped tissue between the black pupil and the white sclera is the iris, and the yellow iris is dark brown. The iris recognition process is as follows:

(1) F (x, A) is related to the value of f (x) on the arc AB, and the parameter λ is the radius of the arc. Use the custom boundary / area detection method to find the inner and outer boundaries of the iris and determine the analysis area;

(2) Establish a grid on the annular analysis area in the following way: Divide the part of the outer circle diameter on the side of the iris (the black line on the left in the figure) into k equal parts, and divide the outer circumference into 1 equal parts (figure (Shown in the middle white line), a k * l grid is established on the iris as a standardized analysis area; (3) Calculate the regularized representation of the texture features ci on the area Ri, ci is obtained from w [m, n] * g [m, n], where w [m, n] is a weight function, and g [m, n ] Is the grayscale of the pixel [m, n];

(4) Perform two or four types of smoothing, keeping the same smoothing order and the same small area (Ri) order for all iris images. In two or four smoothing cases, the code lengths are 2 * K * L and 4 * K * L, and K: and L are determined by k and 1, respectively. Calculate hd and authentication threshold δ for identity authentication.

2. Fingerprint recognition

The fingerprint image is shown in Figure 2. The recognition process is as follows:

(1) With reference to the positioning template provided by the image acquisition device, F (x, λ) is related to the value of f (x) on the ellipse with the focal length as the parameter, and the outer boundary of the fingerprint is obtained using a custom boundary / area detection method . The area within the ellipse shown in the figure is the analysis area;

(2) Establish a grid on the ellipse analysis area in the following way: divide the major axis of the ellipse into k equal parts, divide the short axis into 1 equal parts, and establish a grid on the fingerprint image analysis area as a standard analysis area;

(3) Calculate the regularized representation of the texture features ci on the area Ri, ci is obtained from w [m, n] * g [m, n], where _w [m, _n ] is a weight function and g [m, n ] Is the grayscale of the pixel [m, n];

(4) Perform two or four smoothings, keep the same smoothing order and the same R order for all fingerprint images. In the case of two smoothings and four smoothings, equal length codes can be obtained. Twice as smooth). Calculate hd and authentication threshold δ for identity authentication.

3. Facial recognition

An image of a part of the tissue of the face can be appropriately selected for recognition, as shown in FIG. 3, and the recognition process is as follows:

(l) F (x, λ) is related to the value of f (x) on the arc with the radius of curvature λ as a parameter. The white feature points shown in the figure are obtained using a custom boundary / area detection method. The polygonal area defined by the feature points is the analysis area.

• (2) Establish a grid on the polygon analysis area in the following way: divide the upper horizontal line segments into k equal parts, divide the line segments between the upper and lower feature points in the middle of the image into 1 equal parts, and establish a grid on the face image As a standardized analysis area.

(3), (4) The two parts are the same as fingerprint identification.

The present invention will use another applicant's invention "image microprocessing technology" for human biological feature recognition, so as to give a comprehensive recognition technology of human biological features such as iris, fingerprints, facial tissues and the like. Using this comprehensive identification technology can develop a comprehensive product of human biometric identity authentication. The emergence of this product meets the needs and expectations of society for secure and practical identity authentication products, and meets the requirements for the use of unmanned monitoring, high reliability, fast, automatic, and secure identity authentication in a network environment. , Which can replace the existing Identity authentication means to build a new security barrier for society and provide security protection for families and individuals.

The application fields of human biometric identification products developed by using the present invention can be summarized into three aspects: ① important entrance and exit control (that is, control of tangible doors, or physical channels), ② access control of information systems (that is, control of intangibles) Door, or channel in the logical sense), ③ used in combination with products in related fields. The following table gives some specific application items in the above areas-

Industrial applications

• This method has the following characteristics:

1. Uniformity: General recognition methods and technologies are given to different human biological tissues. Using this method and technology can develop comprehensive human biometric identification products;

2. Reliability: Analyze the image on the micro scale, so that the rate of misunderstanding is far lower than the existing methods and technologies;

3. Practicality: The image acquisition equipment of different human biological tissues can be integrated into a single device, the computer software is integrated into a single chip, and different human biological tissue identification systems are integrated into a single product for use in a unified environment.

Claims

Rights request

1. A method for identifying human biometrics using an image acquisition device and a computer, the steps are:

(1) Image separation

Use an image acquisition device to collect a human biological tissue image, and use a computer to detect an edge, a contour line or a region of a specific analysis object from the human biological tissue image in order to separate and extract the analysis object;

(2) Regional standardized transformation

(3) Regularization of texture features

Regularize the texture in the standard grid with a computer;

(4) Equal-length, maximum entropy coding

(5) Feature code analysis

Use a computer to analyze two different images of the same human biological tissue using the methods in (1)-(4) above to determine whether the two images come from the same human biological tissue of the same person. For identity authentication, if the corresponding code points with different values in the two sets of codes occupy less than or equal to the specified value, the two images are the same, otherwise they are different.

2. A method for identifying human biometrics according to claim 1, wherein the method for image separation is:

The two-dimensional grayscale image of human biological tissue is represented by F (x), where X represents the position of a pixel point in the image, and f (X) is the grayscale of the point; the boundary / region automatic detection mechanism in the recognition process It is called a custom boundary / area detector. The process of customizing the boundary / area detector is as follows:

a. Customize the structure of the boundary or region to be detected: If the boundary to be detected is composed of straight line segments, define a linear boundary detector F (x, A), which is usually an image f (x) with λ Is the line integral or area score of the parameter, F (x, λ) is a function of f (x), and λ is a geometric parameter related to the shape of the boundary or region to be detected;

b. Smooth the detector: determine the scope of the boundary detector F (x, A), and smooth each point on the boundary detector F (x, λ) with a smoothing function on the scope;

c. Derivate λ or calculate the change amount: Calculate the change in the overall characteristics of each possible boundary when λ changes to search for the true boundary from the possible boundaries;

d. Calculate the extreme value of the derivative or change: Determine by calculating the extreme value of the derivative or change Set real boundaries.

The method for identifying human biometrics according to claim 1, characterized in that the method of normalizing the region is all or part of the following methods a-d:

a transform the image of an analysis area on the plane to another area;

b. Transform the boundary of an analysis area from one shape to another shape in the image;

c. Standardize the texture features inside the analysis object;

d. transform the image from one coordinate space to another coordinate space;

In general, it is not required to perform all the transformations in the above requirements, and the general methods of a, b, and c above are now given:

a. Establish a standard grid on a given analysis area. The establishment of the grid must follow two principles: one is to establish the same grid for the same physiological tissue, and the other is to divide the grid in the grid according to the area's geometry; If the geometry of the analysis area is already suitable for analysis, other transformations can be omitted after the standard grid is established;

b. If you want to transform the geometry of the analysis area, you must transform the line segments in the grid to the target area one by one. The transformation is performed in the following way: Let the intersection point of the line segment a and the boundary contour line C1 in the transformation be xl, x2, a (x) are points on line segment a, then the value of the transformed image f (x) in the target region is determined by two known standardized processing functions hl (x, xl), h2 (x, x2) .

4. The method for human biological feature recognition according to claim 1, wherein the method for regularly transforming the texture features is-centered on the geometric center of each small area in the analysis area after the area feature transformation. A coordinate system is established on each small area, and the texture on each small area is regularly transformed in this coordinate system:

a. Smoothing the image: weighting the gray value of each point, these weight coefficients can be selected from the sine function, cosine function, exponential function, hyperbolic function or other functions in addition to the coordinates.

b. For a given point, find the sum of smoothing multiple times with different weight function pairs, that is, multiple weighted summation, generally even numbers are used;

c. Find the weighted sums described in a., b. above for all points in a small area, then the weighted sum is a regularized representation of the texture features of the points on the small area.

The method for identifying human biometrics according to claim 1, wherein the method of equal length and maximum entropy coding is:

a. The weight functions in the regularized transformation of texture features can all be mapped to each other, and the regularized representations of each other's mapping are regarded as the same;

b. In order to meet the above requirements, the invariant in the mapping is taken as the coding threshold, and each cell If the weighting value of each point in the domain is less than or equal to the encoding threshold, the corresponding code point value is 0, otherwise the value is 1;

c. For the determined human biological tissues, the smoothing processing of the same number of times and the same weight function is used in the regular transformation of texture features, thereby obtaining an equal-length encoding. This encoding method is the maximum entropy encoding, which contains the maximum information. the amount. '

The method for identifying human biometrics according to claim 1, wherein the method for analyzing the feature code is:

a. Calculate the percentage of corresponding code points with different values in the two groups of codes in all code points;

b. Determine the authentication threshold based on statistical characteristics. When the above percentage is less than or equal to the authentication threshold, they are considered the same, otherwise they are considered different.