Biological feature template generation and query method of revocable binary feature based on spherical hash

Abstract

The invention discloses a spherical hash-based biological feature template generation and query method for revocable binary features, and the generation method comprises the steps: obtaining a biological feature image, and carrying out the preprocessing of the image, and obtaining a standard biological feature image; extracting a feature vector of the standard biological feature image, generating a Cake-cut matrix and an expanded feature vector based on the extracted feature vector, and generating a transformed feature vector according to the Cake-cut matrix and the expanded feature vector; carrying out random permutation on the transformed vector by adopting a random seed to obtain a permuted feature vector; and generating a binary hash template with the length of C according to the size relationship between the distance between the replaced feature vector and each of the trained C sphere centers and the trained radius corresponding to the sphere center. The feature template generated by the method is high in safety, small in occupied memory and good in revocation and irrelevance.

Description

Technical Field

[0001] The invention belongs to the technical field of biometric identification, and in particular relates to a biometric feature template generation and query method based on spherical hashing and revocable binary features. Background Art

[0002] Thanks to the development of deep learning and computing power, biometric recognition technology has been applied to many aspects of our lives. Among various biometric features, face has become one of the most widely used authentication methods due to its non-invasive, easy to obtain and highly robust characteristics, which has greatly facilitated people's lives. However, due to its wide application, face recognition systems are often used as targets of attack. Existing work has shown that attackers can train specific models to learn the mapping from face feature vectors to face images, thereby attacking face recognition systems and threatening user privacy. Therefore, we need to protect the face templates stored in the face recognition system database, improve the security of the face recognition system, and better protect user privacy.

[0003] Some existing biometric template generation methods have the following problems: 1. They may lead to the leakage of original biological data; 2. They are only applicable to specific modalities and are not compatible with multiple modalities; 3. The computational complexity of generating biometric templates is high and the generation speed is slow; 4. The security of template entropy values ​​is often ignored, thus causing security risks. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a biometric template generation and query method based on spherical hashing and revocable binary features.

[0005] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] The present invention provides a method for generating a biometric template with revocable binary features based on spherical hashing, comprising:

[0007] Acquire a biometric image, and preprocess the biometric image to obtain a standard biometric image;

[0008] Extracting a feature vector of the standard biometric feature image, generating a Cake-cutting matrix and an expanded feature vector based on the extracted feature vector, and generating a transformed feature vector according to the Cake-cutting matrix and the expanded feature vector;

[0009] Using a random seed corresponding to the biometric image, randomly permuting the transformed vector to obtain a permuted feature vector;

[0010] A binary hash template of the biometric image with a length of C is generated according to the distance between the replaced feature vector and each of the C trained sphere centers and the size relationship between the distance and the trained radius corresponding to the sphere center; the C trained sphere centers and the trained radius corresponding to each trained sphere center are obtained by training based on a spherical hash algorithm and multiple transformed feature vectors that correspond one to one to multiple biometric images used for training; C is a positive integer greater than 1.

[0011] The present invention also provides a query method for a biometric template based on revocable binary features, comprising:

[0012] The above-mentioned biometric template generation method based on spherical hashing and revocable binary features is used to generate a binary hash template of the biometric image to be queried, thereby obtaining a query template;

[0013] Calculating the matching degree between the query template and each registered binary hash template; wherein each registered binary hash template is a binary hash template of a biometric feature map for registration generated by the above-mentioned biometric feature template generation method based on spherical hashing and revocable binary features;

[0014] According to the size relationship between the matching degree and a preset threshold, it is determined whether the query template and the registered binary hash template belong to the same biometric feature, and a query result is obtained.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The present invention generates a Cake-cutting matrix and an expanded feature vector based on the feature vector of the biometric image, generates a transformed feature vector based on the Cake-cutting matrix and the expanded feature vector, and generates a feature template based on the transformed feature vector. The original biometric aggregation vector can be projected into a transformation space that has no association with the original biometric information, thereby ensuring the irreversibility of the transformation, thereby not causing the leakage of the original biometric information. Moreover, due to the limitation of current computing resources, even if the generated feature template and the Cake-cutting matrix are lost, the thief cannot calculate the original biometric information within a finite time, so it has good security. The present invention uses a random seed as a user password, randomly permutes the transformed vector, obtains a permuted feature vector, generates a feature template based on the permuted feature vector, and even if the generated feature template is lost, a new random seed can be used to regenerate a new permuted feature vector, so as to generate a new feature template based on the new permuted feature vector, and has good revocability and irrelevance. When generating a feature template, the present invention only involves the calculation of the distance between the data point and the center of the sphere, and the comparison of the distance with the radius corresponding to the center of the sphere. It consumes less time, has low computational complexity, and has high efficiency in generating feature templates. When generating a feature template, the present invention can achieve a good bit distribution balance, that is, the probability of each bit of the hash code being assigned to 0 and 1 is close to 50%, so that the generated feature template has a higher entropy value, thereby further improving security. In addition, the generated feature template is in a binary space, and compared with the original vector, the feature template occupies a smaller storage space, so that the subsequent feature template matching is faster. In addition, the present invention has no special restrictions on the type of biometric features, and can be extended to the template protection of all biometric features, that is, it can be compatible with multiple modalities, thereby having a wider range of applications.

[0017] The query method provided by the present invention uses a biometric feature template based on revocable binary features for matching query, so the feature template occupies a smaller storage space and has higher security, thus achieving fast query speed and high security.

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a method for generating a biometric template with revocable binary features based on spherical hashing provided by an embodiment of the present invention;

[0020] Figure 2 is a schematic diagram of an exemplary method for generating a biometric template and a method for training a sphere center and a radius provided in an embodiment of the present invention;

[0021] Figure 3 It is a flowchart of a method for querying a biometric template with revocable binary features based on spherical hashing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0023] Figure 1 is a flow chart of a method for generating a biometric template with revocable binary features based on spherical hashing provided by an embodiment of the present invention, such as Figure 1 As shown, the method includes:

[0024] S101, acquiring a biometric image, and preprocessing the biometric image to obtain a standard biometric image.

[0025] Here, the biometric image may be any biometric image, for example, a face image, a fingerprint image, or an iris image.

[0026] Here, the acquired biometric image is one or more biometric images, and each image is pre-processed and subsequently processed in S102 to S104.

[0027] Here, a face detection model may be used to first detect key points of a biometric image, and then affine transformation may be applied to align and crop the image to a standard size to obtain a standard biometric image.

[0028] S102, extracting feature vectors of the standard biometric feature image, generating a Cake-cutting matrix and an expanded feature vector based on the extracted feature vector, and generating a transformed feature vector based on the Cake-cutting matrix and the expanded feature vector.

[0029] S103 , using a random seed corresponding to the biometric feature image, randomly permuting the transformed vector to obtain a permuted feature vector.

[0030] S104. Generate a binary hash template of a biometric image with a length of C according to the relationship between the distance between the replaced feature vector and each of the C trained sphere centers and the trained radius corresponding to the sphere center; the C trained sphere centers and the trained radius corresponding to each trained sphere center are obtained by training based on a spherical hash algorithm and a plurality of transformed feature vectors corresponding one to one to a plurality of biometric images used for training; C is a positive integer greater than 1.

[0031] The above S102 is implemented by the following steps:

[0032] S1021. Use a pre-trained feature vector extraction model to extract feature vectors of a standard biometric image to obtain extracted feature vectors.

[0033] For example, the pre-trained feature vector extraction model may be a pre-trained face recognition model ArcFace, and the extracted feature vector may be a feature vector of a specified dimension, such as a feature vector of 512 dimensions.

[0034] S1022. Expand the dimension of the extracted feature vector to obtain an expanded feature vector.

[0035] Here, the dimension of the expanded feature vector is twice the dimension of the extracted feature vector. For example, when the dimension of the extracted feature vector is 512, the order of the extracted 512-dimensional feature vector is reversed to obtain a new 512-dimensional feature vector, and the new 512-dimensional feature vector is concatenated with the 512-dimensional feature vector before the order reversal to obtain a 1024-dimensional expanded feature vector.

[0036] S1023. Generate a Hadamard matrix according to the dimension of the expanded eigenvector, and generate a Cake-cutting matrix according to the Hadamard matrix.

[0037] Here, each row vector of the Hadamard matrix is ​​extracted and converted into a two-dimensional square matrix, and the number of connected regions of the square matrix is ​​calculated. Finally, all rows of the Hadamard matrix are arranged in ascending order according to the number of connected regions of the square matrix corresponding to each row of the Hadamard matrix to obtain a Cake-cutting matrix.

[0038] For example, when the expanded eigenvector is 1*1024, the generated Hadamard matrix is ​​1024*1024, that is, 1024 rows and 1024 columns. Specifically, each row vector of the 1024*1024 Hadamard matrix is ​​extracted and converted into a two-dimensional square matrix, that is, the 1*1024 row vector is converted into a 32*32 square matrix, and the number of connected regions of the 32*32 square matrix is ​​calculated. Finally, all rows are arranged in ascending order according to the number of connected regions of the square matrix corresponding to each row, so as to obtain a 1024*1024 Cake-cutting matrix CC∈R 1024×1024 .

[0039] S1024. Multiply some columns in the Cake-cutting matrix by the expanded feature vector to obtain a transformed feature vector.

[0040] Here, 1 / 2 columns in the Cake-cutting matrix are multiplied by the expanded eigenvector to obtain the transformed eigenvector.

[0041] For example, from CC∈R 1024×1024 Randomly select 512 columns from the matrix to get a partial Cake-cutting matrix Afterwards, And the expanded feature vector Multiply them to get the transformed vector

[0042] For the above S103, the random seed corresponding to each biometric image is randomly generated.

[0043] In the present invention, each sphere center is a vector. Based on this, the above S104 is implemented by the following steps:

[0044] S1041. Calculate the distance between the replaced feature vector and the cth trained sphere center to obtain the cth distance; the value of c ranges from 1 to C.

[0045] S1042. When the cth distance is greater than the trained radius corresponding to the cth trained sphere center, 0 is used as the binary value of the cth bit in the binary hash template of the biometric image; when the cth distance is less than or equal to the trained radius corresponding to the cth trained sphere center, 1 is used as the binary value of the cth bit in the binary hash template of the biometric image, and a binary hash template with a length of C for the biometric image is obtained.

[0046] For example, a binary hash template of length C of a biometric image It can be expressed as:

[0047]

[0048] in, for The binary value of the cth bit of represents the permuted eigenvector and the cth trained ball center P c The Euclidean distance between c represents the cth trained sphere center P c One-to-one correspondence between radii.

[0049] In the present invention, before the above S104, the following steps are further performed to obtain the trained C sphere centers and the trained radius corresponding to each trained sphere center:

[0050] S001. Acquire multiple biometric images for training, and generate multiple transformed feature vectors as multiple data points based on the multiple biometric images for training; the number of the multiple data points is m; and m is a positive integer greater than 1.

[0051] Here, the principle of generating the transformed feature vector of each biometric feature image used for training is the same as S102 above, and will not be repeated here.

[0052] For example, when there are m biometric images for training, the vector set consisting of m extracted feature vectors corresponding to the m biometric images for training is X train ={x train，1 , x train，2 , ..., x train，m}∈R m×512 When , the vector set consisting of m transformed feature vectors corresponding to m biometric images used for training can be expressed as From the Cake-cutting matrix CC∈R 1024×1024 A randomly selected portion of the columns.

[0053] S002. Obtain C initial sphere centers and C initial radii corresponding to each other according to multiple data points, a preset sampling number n and a preset number of sphere centers C; the value of c ranges from 1 to C.

[0054] Specifically, when generating the cth initial center of a sphere, n data points are randomly selected from multiple data points, and the average value of the selected n data points is calculated, and the average value is used as the cth initial center of a sphere; then, the distance between the cth initial center of a sphere and each of the multiple data points is calculated to obtain multiple distances; then, the multiple distances are arranged in ascending order, and a distance that is at the median of the arrangement order is used as the cth initial radius that corresponds one-to-one to the cth initial center of a sphere, and in this way, C initial centers of a sphere and C corresponding initial radii are obtained.

[0055] S003. During the p-th training, calculate the number of data points in two spheres of the C spheres obtained in the p-1th time among the multiple data points, and obtain multiple numbers of one-to-one correspondence between multiple pairs of spheres; wherein each pair of spheres is two spheres; the c-th sphere obtained in the p-1th time is a sphere formed by the c-th center obtained in the p-1th time and the c-th radius obtained in the p-1th time; p is greater than or equal to 1, and when p is 1, the C centers obtained in the p-1th time and the one-to-one corresponding C radii are the C initial centers and the C initial radii.

[0056] S004. When multiple numbers meet the preset convergence condition, the C sphere centers obtained for the p-1th time are used as the trained C sphere centers, and the C radii obtained for the p-1th time are used as the trained C radii that correspond one-to-one to the trained C sphere centers.

[0057] Specifically, when each of the multiple quantities is The average of the absolute values ​​after subtraction is less than or equal to And, the standard deviation of multiple quantities is less than or equal to , indicating that multiple quantities meet the preset convergence conditions, where ε m is the first preset constraint value, ε s is the second preset constraint value; otherwise, it indicates that the multiple quantities do not meet the preset convergence condition. The preset convergence condition is specifically expressed as the following formula:

[0058]

[0059] Among them, i，j Indicates each quantity in the multiple quantities in S003 that correspond to multiple pairs of spheres, std-dev(o i,j ) means o i,j The standard deviation of ε m and ε s It indicates the degree of constraint on the convergence condition. The smaller the value, the more stringent the convergence condition is, and vice versa.

[0060] S005. When multiple quantities do not meet the preset convergence conditions, determine the C centers and C radii obtained for the pth time based on the multiple quantities, m and the C centers obtained for the p-1th time, and perform the p+1th training based on the C centers and C radii obtained for the pth time. The training is iterated in this way until C trained centers and C trained radii that correspond one to one are obtained.

[0061] Specifically, in S005, “determining the C sphere centers and C radii obtained for the pth time according to the multiple numbers, m, and the C sphere centers obtained for the p-1th time” is implemented by the following steps:

[0062] S1. For each pair of spheres in the plurality of pairs of spheres, determine the force between the centers of the pair of spheres according to the number of the pair of spheres that correspond one to one.

[0063] For example, when a pair of spheres is the i-th sphere and the j-th sphere, and the number of the one-to-one correspondence between the pair of spheres is o i,j When , the expressions of the force between the center of the i-th sphere and the center of the j-th sphere are as follows:

[0064]

[0065] f j←i =-f i←j ;

[0066] Among them, f i←j is the center P of the jth sphere j For the center P of the i-th sphere i The force, f j←i is the center P of the i-th sphere i For the center P of the jth spherej The force of action.

[0067] S2. Based on the forces acting on multiple pairs of spheres, determine the sum of the forces acting on the centers of the remaining C-1 spheres on each center of the sphere obtained at the p-1th time, and obtain the resultant force on the center of each sphere obtained at the p-1th time.

[0068] For example, for the i-th sphere mentioned above, the center of each sphere is aligned with the center of the i-th sphere P. i By averaging the forces of the other spheres, we can get the center of the sphere P of the i-th sphere. i The resultant force at the center of the sphere, P i The resultant force f i The expression is:

[0069] S3. For each sphere center obtained for the p-1th time, a new sphere center is generated as a sphere center obtained for the pth time according to the resultant force of the sphere center and the sphere center.

[0070] Specifically, for the center P i For the pth time, the center of the sphere P i The expression is: i p =P i p-1 +f i , P i p-1 is the center of the sphere P obtained for the p-1th time i .

[0071] S4. According to the multiple data points and the C sphere centers obtained for the pth time, determine the C radii that correspond one to one to the C sphere centers obtained for the pth time, and obtain the C radii obtained for the pth time.

[0072] Here, the principle of S4 is the same as the principle of determining the initial radius in the above S002, and will not be repeated here.

[0073] The following Figure 2 , the above-mentioned biometric template generation method based on spherical hashing revocable binary features is exemplified. Figure 2As shown, in the generation stage of the facial feature template of the face image, first, the face image is preprocessed to obtain a standard face image, then the DNN model is used to extract the original feature 1*512 of the standard face image, the extracted original feature 1*512 is inverted to obtain the inverted feature 1*512, the original feature 1*512 and the inverted feature 1*512 are spliced ​​to obtain the aggregated feature 1*1024 (i.e. the expanded feature vector mentioned above), then the aggregated feature 1*1024 is multiplied with the partial CC matrix of 1024*512 (i.e. the partial columns in the above Cake-cutting matrix) to obtain the 1*512 transformation feature (i.e. the above transformed feature vector), the transformation feature is randomly permuted using a random permutation seed to obtain the permuted feature, then the distance between the permuted feature and each trained sphere center is calculated, and the size relationship between the distance and the radius of each sphere center UI shadow is compared, and the hash template of the face image, i.e. the facial feature template, is generated according to the size relationship. Continue to refer to Figure 2 In the training phase of the center and radius of the sphere, m training images are obtained, and the m training images are preprocessed to obtain m standard face images for training. Then, the DNN model is used to extract the original features of each standard face image for training. After that, the same principle as the generation phase is used to determine the aggregate features of each standard face image for training, and the same principle as the generation phase is used to determine the transformation features of each standard face image for training. The m transformation features corresponding to the m standard face images for training are used as m* 512 training data, initialize C different sphere centers P according to the training data, and determine C different radii R corresponding to the C different sphere centers P, then calculate the force between the sphere centers and change the position of the sphere center, update the radius R corresponding to the sphere center, then judge whether the P sphere center after the change of position meets the convergence condition, if not, continue to change the position of the sphere center and update the radius R corresponding to the sphere center, until the P sphere center after the change of position meets the convergence condition, and obtain a stable sphere center P and radius R (that is, the above-mentioned trained C sphere centers and the one-to-one corresponding C radii).

[0074] The present invention has the following advantages and beneficial technical effects:

[0075] (1) The vector transformation algorithm based on the partial cake-cutting matrix proposed in the present invention projects the original biometric aggregation vector into a transformation space that has nothing to do with the original biometric information, thereby ensuring the irreversibility of the entire system. Limited by current computing resources, even if the generated feature template and the cake-cutting matrix are lost, the thief cannot calculate the original biometric in a finite time, so the system has good security.

[0076] (2) The present invention uses a random seed as a user password, randomly permutes the transformed vector, obtains a permuted feature vector, and generates a feature template based on the permuted feature vector. Even if the generated feature template is lost, a new random seed can be used to regenerate a new permuted feature vector, and a new feature template can be generated based on the new permuted feature vector, which has good revocability and irrelevance.

[0077] (3) The present invention can achieve a better bit distribution balance by virtue of the characteristics of the spherical hash algorithm, that is, the probability of each bit of the hash code being assigned to 0 and 1 is close to 50%, so that the generated biometric template has a higher entropy value, thereby further improving the security of the system. In addition, the biometric template generated by the system is a hash template, and the hash template is in a binary space. Compared with the original vector, the hash template occupies a smaller storage space (a single hash template only needs 1KB to store), so the matching speed is faster.

[0078] (4) Compared with other methods, this method only involves calculating the Euclidean distance between the data point and the center of the sphere, and comparing the distance with the radius of the sphere when generating the hash template, which consumes less time. In addition, the present invention has no special restrictions on the type of biometrics and can be extended to the template protection of other biometrics.

[0079] (5) This invention can effectively protect original facial information from being illegally stolen, promote the safe development of the information industry, and has important market value.

[0080] The present invention also provides a query method based on a biometric template with revocable binary features, such as Figure 3 As shown, the method includes:

[0081] S201. A biometric template generation method based on spherical hashing and revocable binary features is used to generate a binary hash template of a biometric image to be queried, thereby obtaining a query template.

[0082] S202, calculating the matching degree between the query template and each registered binary hash template; wherein each registered binary hash template is a binary hash template of a biometric feature map for registration generated by a biometric feature template generation method based on spherical hashing and revocable binary features.

[0083] S203: Determine whether the query template and the registered binary hash template belong to the same biometric feature based on the size relationship between the matching degree and the preset threshold, and obtain a query result.

[0084] Specifically, the expression of matching degree is as follows:

[0085]

[0086] Among them, S(H enroll ,H query ) is the matching degree, H enroll is a registered binary hash template, H query is the query template, TN=C represents the template length, XOR represents the exclusive OR operation, h enroll,i and h query,i is the hash value at the i-th position in the registration template and query template.

[0087] The query method provided by the present invention uses a biometric feature template based on revocable binary features for matching query, so the feature template occupies a smaller storage space and has higher security, thus achieving fast query speed and high security.

[0088] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0089] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0090] In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good effects.

[0091] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for generating a biometric template with revocable binary features based on spherical hashing, characterized in that: include: Acquire a biometric image, and preprocess the biometric image to obtain a standard biometric image; Extracting a feature vector of the standard biometric feature image, generating a Cake-cutting matrix and an expanded feature vector based on the extracted feature vector, and generating a transformed feature vector according to the Cake-cutting matrix and the expanded feature vector; Using a random seed corresponding to the biometric image, randomly permuting the transformed vector to obtain a permuted feature vector; A binary hash template of the biometric image with a length of C is generated according to the distance between the replaced feature vector and each of the C trained sphere centers and the size relationship between the distance and the trained radius corresponding to the sphere center; the C trained sphere centers and the trained radius corresponding to each trained sphere center are obtained by training based on a spherical hash algorithm and multiple transformed feature vectors that correspond one to one to multiple biometric images used for training; C is a positive integer greater than 1.

2. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 1, characterized in that: Each sphere center is a vector; the binary hash template of the biometric image with a length of C is generated according to the distance between the replaced feature vector and each of the C sphere centers trained, and the size relationship between the trained radius corresponding to the sphere center, including: Calculate the distance between the replaced feature vector and the cth trained sphere center to obtain the cth distance; the value of c ranges from 1 to C; When the c-th distance is greater than the trained radius corresponding to the c-th trained sphere center, 0 is used as the binary value of the c-th bit in the binary hash template of the biometric image; when the c-th distance is less than or equal to the trained radius corresponding to the c-th trained sphere center, 1 is used as the binary value of the c-th bit in the binary hash template of the biometric image to obtain a binary hash template with a length of C for the biometric image.

3. The method for generating a biometric template based on spherical hashing and revocable binary features according to claim 1, characterized in that: The step of extracting a feature vector of the standard biometric feature image, generating a Cake-cutting matrix and an expanded feature vector based on the extracted feature vector, and generating a transformed feature vector according to the Cake-cutting matrix and the expanded feature vector includes: Using a pre-trained feature vector extraction model to extract the feature vector of the standard biometric image, to obtain the extracted feature vector; Expanding the dimension of the extracted feature vector to obtain an expanded feature vector; Generate a Hadamard matrix according to the dimension of the expanded eigenvector, and generate a Cake-cutting matrix according to the Hadamard matrix; Multiplying some columns in the Cake-cutting matrix by the expanded feature vector to obtain the transformed feature vector.

4. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 1, characterized in that: Before generating a binary hash template of the biometric image with a length of C according to the relationship between the distance between the replaced feature vector and each of the C trained sphere centers and the trained radius corresponding to the sphere center, the method further includes: Acquire the plurality of biometric images for training, and generate the plurality of transformed feature vectors as a plurality of data points according to the plurality of biometric images for training; the number of the plurality of data points is m; m is a positive integer greater than 1; According to the plurality of data points, the preset sampling number n and the preset number of sphere centers C, C initial sphere centers and C initial radii corresponding to each other are obtained; the value of c ranges from 1 to C; During the p-th training, the number of data points in the multiple data points that are simultaneously in two spheres of the C spheres obtained in the p-1th time is calculated to obtain multiple numbers of one-to-one correspondence between multiple pairs of spheres; wherein each pair of spheres is two spheres; the c-th sphere obtained in the p-1th time is a sphere formed by the c-th sphere center obtained in the p-1th time and the c-th radius obtained in the p-1th time; p is greater than or equal to 1, and when p is 1, the C sphere centers obtained in the p-1th time and the one-to-one corresponding C radii are the C initial sphere centers and the C initial radii; When the multiple numbers meet the preset convergence condition, the C sphere centers obtained for the p-1th time are used as the trained C sphere centers, and the C radii obtained for the p-1th time are used as the trained C radii that correspond one-to-one to the trained C sphere centers; When the multiple numbers do not meet the preset convergence condition, the C centers of the sphere and the C radii obtained for the p-1th time are determined according to the multiple numbers, m and the C centers of the sphere obtained for the p-1th time, and the p+1th training is performed according to the C centers of the sphere and the C radii obtained for the pth time, and the training is iterated in this way until the trained C centers of the sphere and the one-to-one trained C radii are obtained.

5. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 4, characterized in that: The step of obtaining C initial sphere centers and C initial radii corresponding to each other according to the plurality of data points, the preset sampling number n and the preset number of sphere centers C comprises: When generating the cth initial sphere center, randomly selecting n data points from the multiple data points, calculating the average value of the selected n data points, and using the average value as the cth initial sphere center; Calculating the distance between the cth initial sphere center and each of the multiple data points to obtain multiple distances; The multiple distances are arranged in ascending order, and a distance in the median of the arrangement order is used as the cth initial radius corresponding to the cth initial sphere center.

6. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 4, characterized in that: When each of the plurality of quantities is The average of the absolute values ​​after subtraction is less than or equal to Furthermore, the standard deviation of the plurality of quantities is less than or equal to , indicating that the multiple quantities meet the preset convergence conditions, where ε m is the first preset constraint value, ε s is a second preset constraint value; otherwise, it indicates that the multiple quantities do not satisfy the preset convergence condition.

7. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 4, characterized in that: The determining the C sphere centers and C radii obtained for the pth time according to the multiple numbers, m, and the C sphere centers obtained for the p-1th time includes: For each pair of spheres in the plurality of pairs of spheres, determining the force between the centers of the pair of spheres according to the number of the pair of spheres in one-to-one correspondence; According to the forces acting on the multiple pairs of spheres, determine the sum of the forces acting on the centers of the remaining C-1 spheres on each sphere center obtained for the p-1th time, and obtain the resultant force on each sphere center obtained for the p-1th time; For each sphere center obtained at the p-1th time, a new sphere center is generated as a sphere center obtained at the pth time according to the resultant force of the sphere center and the sphere center; According to the multiple data points and the C sphere centers obtained for the pth time, C radii corresponding to the C sphere centers obtained for the pth time are determined to obtain the C radii obtained for the pth time.

8. The method for generating a biometric template with revocable binary features based on spherical hashing according to claim 7, characterized in that: When a pair of spheres is the i-th sphere and the j-th sphere, and the number of the one-to-one correspondence between the pair of spheres is o i,j When , the expressions of the force between the center of the i-th sphere and the center of the j-th sphere are as follows: f j←i =-f i←j ; Among them, f i←j is the center P of the jth sphere j For the center P of the i-th sphere i The force, f j←i is the center P of the i-th sphere i For the center P of the j-th sphere j The force of action.

9. A query method for a biometric template based on revocable binary features, characterized in that: include: Using the method described in any one of claims 1 to 8 above, a binary hash template of the biometric image to be queried is generated to obtain a query template; Calculating the degree of match between the query template and each registered binary hash template; wherein each registered binary hash template is a binary hash template of a biometric feature map for registration generated by the method described in any one of claims 1 to 8 above; According to the size relationship between the matching degree and a preset threshold, it is determined whether the query template and the registered binary hash template belong to the same biometric feature, and a query result is obtained.

10. The query method of biometric template based on revocable binary features according to claim 9, characterized in that: The expression of the matching degree is as follows: Among them, S(H enroll ,H query ) is the matching degree, H enroll is the registered binary hash template, H query is the query template, TN=C represents the template length, XOR represents the exclusive OR operation, h enroll,i and h query,i is the hash value at the i-th position in the registration template and the query template.

Images