CN101246543B - Examiner identity identification method based on bionic and biological characteristic recognition - Google Patents

Abstract

The invention discloses an examinee identity authentication system based on bionics and biometric identification, which comprehensively utilizes various biometric identification methods based on high-dimensional geometric shape adaptive coverage theory to realize the identification of the examinee's identity. First, the test taker's pen fingerprints, online signatures and face images are obtained through the collection equipment, and then these data are extracted and mapped into points to be observed in the corresponding high-dimensional space. Finally, based on similar sample points, the points are mapped in the high-dimensional space. Continuity, the network matching degree of different biometric characteristics is obtained through the relationship between the points to be observed and the coverage area of the sample set, and then the identity of the examinee is identified through the matching degree fusion decision-making algorithm, and the sample set is realized through the automatic addition of new verification data. Dynamic updates and trend predictions. The invention is quick in identification and has accurate results. It is not only suitable for the identification of examinees under the existing examination mode, but also has broader application prospects in the future computer-based examination mode.

Description

Examiner identification method based on bionics and biometric identification

technical field

The invention relates to an identity identification and authentication system, in particular to an identity identification system which integrates multiple biological features and is based on bionic pattern recognition.

Background technique

Examination is a concentrated review of the examinee's knowledge mastering ability and psychological quality, and it is the only way for every student to pass. At the same time, exams mean opportunities, and an important exam like the college entrance examination will have a profound impact on the life of the examinee. However, in recent years, the examination order of various social examinations has been severely challenged, and the phenomenon of substitute examinations has emerged in an endless stream, seriously undermining the seriousness and fairness of the examinations. The reasons are, on the one hand, due to weak administrative supervision, and on the other hand, because the existing examination model is outdated and technical means are backward, making it difficult to conduct accurate qualification examinations for candidates. Taking the college entrance examination as an example, the invigilators mainly confirm the identity of the candidates by comparing the photos on the admission ticket, the photo on the ID card, and the candidates themselves. Muddle through.

With the development and improvement of the pattern recognition theory, the identification technology based on the biological and behavioral characteristics of the human body has made great progress. Face recognition, fingerprint recognition, iris recognition, signature recognition, gait recognition, etc. have made great progress in their respective research fields The corresponding results were obtained and gradually applied to actual production. These biological characteristics not only reflect the static image information (face, fingerprint, iris) of the person, but also contain the dynamic behavior information of the person (online signature, gait). The accuracy of biometric authentication and identification, user acceptance and cost are different, and each has its own advantages and disadvantages. For users, identification and authentication through the face system is the most friendly way, while iris recognition and authentication have already It has been proved to be the most reliable, stable and accurate detection method. The online signature and handwriting recognition system is also widely accepted by users because of its convenient collection and simple operation. However, these commonly used recognition systems face many problems when they are used alone. For example, the face recognition system is very sensitive to factors such as illumination, posture, and expression. In actual use, it is likely that the quality of the iris sample collected by the user is too poor or the user suffers from ophthalmological diseases, etc., the signature and handwriting recognition, for the same user, the signature and handwriting in different periods and There will also be large differences in different states, not to mention the problems of counterfeiting and counterfeiting. The above-mentioned problems can be effectively solved by integrating multiple biometric identification authentication systems, which can be used for accurate identification of the examiner's identity, and technically prevent the use of forged documents to deceive the invigilators.

With regard to identification methods for fusion of various biometric features, some well-formed research results have obtained patent authorization. For example, a Chinese patent with publication number CN1304114A discloses an identification fusion method based on multiple biometric features, which belongs to pattern recognition. field. It uses human biological characteristics, such as: face, iris, fingerprints, handwriting, etc., to identify people, and uses the standard normalization method to normalize all feature outputs to the same range, and then adopts automatic Organizational feature mapping neural network and fuzzy neural network technology and other methods are used for fusion; the Chinese patent with the publication number CN1794266A discloses an identity recognition and authentication method for biometric fusion, which is characterized in that firstly, the user's face, The biometric features of iris, online signature and offline handwriting are then sent to the corresponding identification and authentication sub-modules for feature extraction and template matching, and the scores obtained after matching are output. After normalization, these scores are either sent to the recognition fusion module, and the final recognition result is obtained through steps such as confidence integration; or sent to the authentication fusion module, mapped to a multidimensional space and classified by a classifier to obtain the final authentication result; or perform authentication fusion again after identification fusion to obtain the final identification result after authentication. After fusion, whether it is verification or identification, the total error rate is lower than that of a single biometric authentication system. These two patents start from the perspective of method research, using various acquisition devices to collect all biometric information, without considering the applicability in practical applications and the intrusion to users, and the main use is still traditional pattern recognition Methods, focusing on the fusion of methods. The Chinese patent with the publication number CN1464478A discloses a non-hyperspherical geometric body covering method in pattern recognition, which includes the following steps: (1) initializing the sample space, dividing the sample space into known sample subspace and unknown sample subspace (2) Start training for a certain type of sample; (3) Construct the relationship between samples of the same type according to the rules, and construct the sample subspace; (4) Use non-hyperspherical geometry for each type of sample (5) Form a closed sample subspace. This invention proposes the best coverage idea for the distribution of similar samples in the feature space, which effectively solves the problems existing in the partition theory in traditional pattern recognition. for further research.

Contents of the invention

The purpose of the present invention is to provide an examiner identification system based on bionics and biometric identification. Through the improvement of the method, the rapid and accurate identification of the examiner's identity can be realized, so as to be applicable to the examiners under the existing paper-and-pencil examination mode. Identity verification is also suitable for identity verification in the future computer-based test mode.

In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: an examiner identification system based on bionics and biometric feature recognition, including using a collector to collect various biometric features, processing the collected data, and performing tests according to the training sample set. Certification identification, specifically includes the following steps:

(1) Simultaneously collect multiple biometric features of the examinee, and transmit the collected data to the front-end processing device after being encrypted by hardware. The biometric features at least include online signature, pen-holding fingerprint pattern and face image;

(2) Preprocessing of the collected data, including filtering and regularizing the data;

(3) Perform feature extraction and feature combination from the pre-processed data, then softly encrypt the data, and then transmit it to the back-end processing equipment through the network;

(4) Back-end analysis and processing of data: the back-end processing equipment decrypts the data packets received from the network, and adopts the biometric identification method based on the adaptive coverage theory of high-dimensional spatial geometry to realize the identification of the examiner;

(5) Dynamic update of the sample set: add the verified data to the sample set, dynamically generate the corresponding template, and make the new template reflect the trend of the sample change.

In the above-mentioned technical scheme, the acquisition method of the step (1) is to first collect the online signature of the examiner through the data board, and use a camera to obtain a relatively clear image of the face and signature posture by using the characteristics that the person's posture is relatively stable when signing. At the same time, the fingerprint image of the pen holding the pen is collected by the signature pen equipped with a multi-functional sensor at the front end. The fingerprint image of the pen holding the pen is all contact information between the finger and the pen when the person holds the pen, including all fingerprint texture fragments and relative positions at the contact.

Wherein, the feature of holding the pen pressure is collected while signing the signature. The method is to set a pressure sensor at each contact of the signature pen, collect the change of the pen holding pressure during the signature process and record it corresponding to the corresponding area of the pen holding fingerprint image .

The synchronous collection method of the above step (1) can be called a "one-stop" method of collecting biometrics, which realizes the collection of as many static (face, fingerprint) and dynamic (signature ) biometric information, which is non-invasive and easy to accept, especially the pressure change of the pen when holding the fingerprint image and signature is a unique biometric characteristic of each person, which is exclusive and exclusive. The combination of other biometric features can significantly improve the accuracy of identification.

Due to the particularity of biometric information, data security issues are particularly prominent. The system first converts the collected data into data format through the encryption chip, and then transmits it to the front-end processing equipment, which eliminates the potential safety hazard in the data transmission process.

Step (2) in the above technical solution preprocesses the collected data, its main purpose is to eliminate irrelevant information in the image, restore useful real information, enhance the detectability of relevant information and simplify the data to the greatest extent, thereby improving The reliability of subsequent processing results. The following preprocessing methods are mainly used in this system: normalization, smoothing, restoration and enhancement.

Step (3) in the above-mentioned technical scheme carries out feature extraction to the preprocessed data, including feature extraction of online signature, face detection and feature extraction and feature extraction of pen-holding fingerprint; It has always been a difficult point in the field of pattern recognition to select useful features from features and organically fuse them to form corresponding feature vectors. Starting from the perspective of bionics, this technical solution adopts the idea of multi-scale analysis to realize the organic fusion between the associated feature sets. Firstly, the extracted features are scaled and divided into overall information and detailed information. The relevance of the algorithm is further divided, and the features with strong correlation are divided into the same associated feature set, and finally the feature fusion is performed to synthesize the associated features into a feature vector.

After feature extraction is performed on face images, online signatures and pen-holding fingerprint images, the front-end processing equipment needs to transmit the merged feature vectors to the back-end processing equipment through the network for analysis and identification. The feature vector contains the biometric information of the examinee, and is an important data source for back-end analysis and identification and update of the training sample set, and its importance is self-evident. The invention adopts an encryption algorithm, a multi-step encryption algorithm to encrypt data, change the data format of the feature vector, and then transmit the data to the back-end processing equipment through the network in the form of data packets, effectively protecting the personal information of the examiners Safety.

After the back-end processing device decrypts the data packet received from the network, it performs authentication and appraisal. The present invention uses different methods to calculate the matching degree for the global feature vector and the detail feature vector. For the global feature vector, it is directly mapped to a point in a high-dimensional space. According to the continuity of similar sample points in a high-dimensional space, the training sample set should be covered by a limited and continuous area. By analyzing the points to be observed and The matching degree between the biometric sample and the training sample set can be obtained by the relationship between the coverage area of the training sample set; and for the detailed features, the dynamic programming method is used for template matching. Then multiply the results of the two to get a matching degree of biological characteristics, and finally use the fusion decision-making algorithm to process the matching degree between various biological characteristic samples and the training sample set, and then weight and sum the matching degrees of various biological characteristics The overall matching degree is obtained, and after comparing with the threshold value, the recognition conclusion can be drawn, and the identification of the examiner's identity can be realized.

"Continuity of similar sample points in high-dimensional space" is the prior knowledge of the distribution of sample points, and it is believed that certain biometric (signature, fingerprint, face) samples from the same person should be continuous in high-dimensional space , that is, there is no mutation between samples, and there is a continuously changing curve between any two sample points, so that one sample point smoothly transitions to another sample point. This is the principle of "homologous continuity", that is, assuming that all the things belonging to class A in the feature space Rn are set A, if there are any two elements x and y in set A, then for any value ε greater than 0, There must exist a set B such that:

$\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; |</span>\left.\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; ,</span>\\ \mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0,1</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; N</span>}\end{array}\right\}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; \&Subset;</span>\mathrm{<span\; class="notranslate">\; A</span>}$

Corresponding to the learning process, it is to select one or more suitable closed surfaces for the distribution of training samples of similar things in the feature space to form a continuous complex geometric shape in a high-dimensional space to reasonably cover the training samples. And this complex geometric shape needs to be adaptively selected according to the distribution of training samples to cover the training sample set in the most reasonable way.

After the user passes the identity verification, the new biometric sample information is automatically recorded by the system. When the storage space is limited, some old information must be dynamically deleted while adding new information to the training sample set. It is meaningful to let the training sample set reflect its changing trend in a dynamically updated manner.

The way to update the training sample set can be considered through the selection of the coverage area of the geometric shape in the high-dimensional space. The present invention realizes the self-adaptive dynamic generation of the coverage area, that is, it is considered that the dynamically updated training sample set should realize the most reasonable dynamic coverage of all samples in the high-dimensional space, and the newly verified biological features are updated to the training sample set, At the same time stale information is filtered out from the coverage area.

Due to the use of the above-mentioned technical solutions, the present invention has the following advantages compared with the prior art:

1. When collecting the biometrics of the test takers, online signatures, faces, and signature postures are collected. A signature pen with a multi-functional sensor is also used to collect all fingerprints in contact with the pen and the corresponding pressure changes when holding the pen. , the rich biological information is exclusive and exclusive, and the combination can significantly improve the accuracy of identification.

2. When the data is transmitted from the acquisition device to the front-end processing device, it is hard-encrypted - encryption chip encryption, and when it is transmitted from the front-end processing device to the back-end processing device, it is encrypted by a multi-step encryption algorithm to ensure data security.

3. The whole system incorporates the idea of bionics, and imitates the way people perceive the world for pattern recognition. Its core ideas are reflected in the following three aspects:

(1) Human thinking can be divided into logical thinking and image thinking. The logical thinking of the brain is developed through the logical reasoning process based on deduction, while the image thinking is the experience summarization process based on induction. developed. Pattern recognition is more of an image thinking problem, which often involves a large number of variables, and it is unrealistic to use multivariable function modeling calculations for strict logical reasoning. Therefore, to solve the problem of pattern recognition, we must find a description method of imagery thinking, and the concept of graphics, that is, the description of geometric shapes in high-dimensional space, is just an effective description method of imagery thinking.

(2) People have the concept of scale change when they identify. Taking signatures, faces, and fingerprints as examples, people not only observe their overall outlines, but also pay attention to their detailed features when identifying them. Therefore, when extracting signature, face, and fingerprint features, not only global information such as aspect ratio and shape must be considered, but also detailed information such as stroke intersections, eye corner lines, etc., to accurately evaluate the recognition objects from different scales.

(3) The memory bank used by people for identification is constantly updated. When a person recognizes an object based on prior knowledge, the existing characteristics of the object will be extracted into information for storage and memory, so as to facilitate accurate identification in the future. Therefore, after comparing the sample to be tested with the existing training sample set and confirming that it is true, it is necessary to automatically add new data that has passed the verification to the sample set, and re-evaluate it, according to each sample in the high-dimensional space In order to remove the old samples reasonably, the new training sample set can reflect the change trend of the recognized object.

4. Starting from the idea of bionics, the system of the present invention comprehensively uses various biometric identification methods based on the adaptive coverage theory of high-dimensional spatial geometry to realize the identification of the examiner's identity. The identification is fast and the result is accurate. It is not only applicable to current There is an examiner's identification in the paper-and-pencil test mode, and it has a broader application prospect in the future computer test mode.

Description of drawings

Fig. 1 is a schematic flow chart of the system of Embodiment 1;

Fig. 2 is the schematic diagram of the collection of fingerprints by holding a pen in Embodiment 1;

FIG. 3 is a schematic diagram of online signature parameters in Embodiment 1;

Fig. 4 is a schematic diagram of homologous continuity in Example 1.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing and embodiment:

Embodiment 1: Referring to accompanying drawings 1 to 4, an examiner identification system based on bionics and biometric identification is elaborated through the following steps:

The first step is the collection of online signature, pen-holding fingerprint pattern and face image data.

Using the method shown in Figure 1, use the data board to collect the online signature of the examinee, and at the same time use the camera to collect face images and pen-holding fingerprint images, and use the multi-function sensor at the front of the special signature pen to collect the pen-holding fingerprint image and when writing. The sensor can not only collect the pen-holding fingerprint image when the person signs, but also can perceive the change of the pen-holding pressure during the signature process. The pen-holding fingerprint image is different from the single complete fingerprint image collected by traditional fingerprint collection equipment. Fingerprint image, which records all the contact information between the finger and the pen when the person holds the pen, including the fingerprint texture fragments and relative positions of all contact points; at the same time, the pressure changes of each contact point during the signing process are also collected by the sensor and corresponding in the corresponding area of the pen-holding fingerprint image.

Subsequently, the system converts the collected data into a data format through a dedicated encryption chip, and then transmits it to the front-end processing equipment, effectively ensuring the data security of the user's biometric information.

The second step is to preprocess the collected data.

The main purpose of preprocessing is to eliminate irrelevant information in the image, restore useful real information, enhance the detectability of relevant information and simplify data to the greatest extent, thereby improving the reliability of subsequent processing results. The following preprocessing methods are mainly used in this system:

(1) Normalization

An image standard form that makes certain features of an image invariant under a given transformation. Certain properties of an image, such as the area and perimeter of an object, are inherently invariant to coordinate rotations. In general, the influence of certain factors or transformations on some properties of the image can be eliminated or weakened through normalization, so it can be selected as the basis for measuring the image. Grayscale normalization, geometric normalization and transformation normalization are three normalization methods to obtain the invariant properties of images.

(2) smooth

A technique for removing random noise from an image. The basic requirement for smoothing technology is to eliminate noise without blurring image contours or lines. The commonly used smoothing methods are median method, local averaging method and k-nearest neighbor averaging method. The size of the local area can be fixed, or it can change with the size of the gray value point by point. In addition, spatial frequency domain bandpass filtering methods are sometimes applied.

(3) Restoration

Correct the image degradation caused by various reasons, so that the reconstructed or estimated image is as close as possible to the ideal image field without degradation. Image degradation often occurs in practical applications, such as aberrations in the optical system, relative motion between the camera and the object will degrade the face image. The basic restoration technique is to regard the acquired degraded image g(x, y) as the convolution of the degraded function h(x, y) and the ideal image f(x, y). There is a relationship G(u, v)=H(u, v)F(u, v) in their Fourier transform. After the degradation function is determined according to the degradation mechanism, F(u, v) can be obtained from this relationship, and then f(x, y) can be obtained by inverse Fourier transform. Usually, M(u, v)=1/H(u, v) is called an inverse filter. In practical applications, since H(u, v) decreases rapidly as the distance from the origin of the uv plane increases, in order to avoid the enhancement of noise in the high-frequency range, when u ² +v ² is greater than a certain limit value W0, M( u, v) equal to 1. The selection of W0 should make H(u, v) have no zero in the range of u ² +v ² ≤ W0. The algebraic method of image restoration is based on the optimal criterion of least squares. Find an estimate that minimizes the value of the goodness criterion function. This method is relatively simple, and the least square method Wiener filter can be derived. When noise is absent, the Wiener filter becomes an ideal inverse filter.

(4) enhanced

Selectively enhance and suppress the information in the image to improve the visual effect of the image, or transform the image into a form more suitable for machine processing, so as to facilitate data extraction or recognition. For example, an image enhancement system can use a high-pass filter to highlight the contour lines of the image, allowing the machine to measure the shape and circumference of the contour lines. There are many methods of image enhancement technology, such as contrast stretching, logarithmic transformation, density layering and histogram equalization, etc. can be used to change the gray tone of the image and highlight the details. In actual application, different methods are often used, and repeated experiments can achieve satisfactory results.

The third step is to extract the features of the preprocessed data and combine them into feature vectors, and then transmit the data to the back-end processing equipment through the network after soft encryption.

(1) Feature extraction of online signature

The online signature contains a wealth of feature information, which can be classified according to the time-space domain information (as shown in Figure 3) and the transformation domain information in the following table.

In addition, some global features in online signatures are also worth paying attention to, such as the signature completion time, the aspect ratio of the signature image, the proportion of the pen uptime to the total time during the signature process, and so on.

For the time-varying features such as speed, inclination, pressure, etc., the present invention first merges them into feature vectors, and then arranges them into feature vector chains, and uses dynamic programming to perform template matching; and for global features, the present invention arranges them into The eigenvectors are then mapped to the points to be observed in the high-dimensional space, and the matching degree is determined by using the relationship between the points to be observed in the high-dimensional space and the sample coverage area.

(2) Face detection and feature extraction

The present invention is based on the skin color model, directly adopts the YCbCr color space and judges the skin color area through the threshold value, and for each point (x, y) in the image, we use f(x, y) to indicate whether the point belongs to the skin color pixel, and obtain The following formula:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; if</span>}\mathrm{<span\; class="notranslate">\; 77</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; Cb</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 127</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; and</span>}\mathrm{<span\; class="notranslate">\; 133</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; Cr</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; 173</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; else</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Then, the particle swarm optimization algorithm is used to segment the skin color area. Particle swarm optimization algorithm is a group-based evolutionary algorithm, and its idea comes from the theory of artificial life and evolutionary computation. When PSO solves an optimization problem, the solution of the problem corresponds to a "particle" or "agent" in the search space. Each particle has its own position and velocity (determining the direction and distance of flight), and a fitness value determined by the optimized function. Each particle remembers, follows the current optimal particle, and searches in the solution space. The process of each iteration is not completely random. If a better solution is found, it will be used as a basis to find the next solution.

Let the PSO be initialized as a group of random particles (random solutions). In each iteration, the particles update themselves by tracking two "extreme values": the first one is the best solution found by the particle itself, called the extreme point (indicated by pbest represents its position), and another extreme point in PSO is the best solution found by the entire population at present, which is called the global extreme point (gbest represents its position).

After finding the two optimal values, the particle updates its velocity and position by the following formula.

V _i ＝ω×V _i +c ₁ ×rand()×(pbest _{i -xi} )+c ₂ ×rand()×(gbest _{i -xi} ) (2)

x _i = x _i +V _i (3)

In the above two formulas:

i=1, 2, ..., M, M is the total number of particles in the group;

V _i is the velocity of the particle;

rand() is a random number between (0, 1);

x _i is the current position of the particle;

c ₁ and c ₂ are learning factors;

ω is non-negative and is called the inertia factor.

In each dimension, the particle has a maximum velocity limit V _max , if the velocity of a certain dimension exceeds the set V _max , then the velocity of this dimension is limited to V _max (V _max >0).

In formula (2), if the value of ω is large, the global optimization ability is strong, but the local optimization ability is weak; if the value of ω is small, the opposite is true. Initially, ω is taken as a constant, and later experiments found that dynamic ω can obtain better optimization results than fixed values. The dynamic ω can change linearly during the PSO search process, or it can change dynamically according to a measure function of the PSO performance. The present invention adopts a linear decreasing weight strategy.

ω ^(t) = (ω _ini -ω _end )(G _k -g)/G _k +ω _end (4)

G _k is the maximum evolution algebra, ω _ini is the initial inertia weight, and ω _end is the inertia weight when iterating to the maximum algebra.

When the skin color area is determined, the segmented area may include some non-face areas, so it is necessary to use the fuzzy membership function to distinguish the face. The fuzzy membership function is defined as follows:

(5)

(5)

This system uses 3 features to describe the human face. are the height of the face, the ratio of height to width, and the vertical offset angle of the face, these parameters are fuzzified by the fuzzy membership function to obtain the fuzzy membership functions μ ₁ (x), μ ₂ (x) and μ ₃ (x), take its product as the total membership function to judge whether the area is a face, that is:

μ(x)=μ ₁ (x)·μ ₂ (x)·μ ₃ (x) (6)

If the membership function value of a certain area is greater than a certain threshold, the area is considered to be a face area. For the single-face application of this system, the area with the largest membership function value is the face area.

After the face area is detected, it is necessary to extract face features, which mainly include changes in the position, size, and shape of eyes, eyebrows, and mouth. First, use the combination of principal component analysis and the proportion characteristics of the face to obtain the initial location of the eyes, and use edge detection and other technologies to quickly and accurately determine the exact location of the black eyes, and then use the proportion features of the face and the color features of the eyebrows and mouth. Eyebrows and mouth can be extracted very accurately. After the salient feature points or attributes of these facial features are extracted, they can be used for face recognition.

(3) Feature extraction of pen-holding fingerprints

As shown in Figure 2, the pen-holding fingerprint collected in the present invention is different from the traditional single complete fingerprint, which records all the contact information between the user's finger and the pen when signing, including the fingerprint texture of all contact points Fragments and their relative positions. Therefore, traditional methods and new methods are integrated in its feature extraction process.

Fingerprints generally have two structures: Henry fingerprint mode and Galton feature. The Henry classification is a standard qualitative design that characterizes the overall structure of peak patterns and is often used to divide fingerprint data. Galton generally defines four characteristics: (1) the start and end of the crest; (2) branches; (3) isolated regions; (4) interference. In order to extract features from digital fingerprint images, eight features including four Galton features are first selected and divided into two categories: original and synthetic. The original features are points, crest endpoints, and branches, respectively. On this basis, synthetic features are defined: isolated regions, burrs, crossovers, bridges and short peaks. In addition, the following features can also be extracted from the pen-holding fingerprint image: the number of fingerprint fragments, the relative position of each fragment, etc.

(4) Feature vector synthesis

After the feature extraction stage, how to select useful features from various features and organically fuse them to form corresponding feature vectors has always been a difficult point in the field of pattern recognition. From the perspective of bionics, the present invention adopts the idea of multi-scale analysis to realize the organic fusion between associated feature sets.

First, scale each feature and divide it into overall information and detail information. Furthermore, according to the correlation between features, it can be further divided. For example, the strength change of the pen grip point when signing, the pressure information recorded on the digital board, the writing speed of the signature, and the change of the stroke direction angle have a strong correlation. into the same associated feature set. Finally, feature fusion is performed, and the associated features are synthesized into feature vectors.

Taking the previous discussion as an example, let d _i (i is the number of points of effort) represent the strength of the pen, p represents the pressure recorded on the digital board, v represents the writing speed of the signature, and θ represents the direction angle, then an associated feature vector can be obtained φ(d _i , p, v, θ).

(5) Data soft encryption and network transmission

Due to the particularity of biometric information, data security issues are particularly prominent. Considering the information security in the network transmission, this system adopts a multi-step encryption algorithm to encrypt the transmission data.

The algorithm uses a sequence of numbers (say, a 128-bit key) to generate a repeatable but highly randomized pseudo-random sequence of numbers. Use 256 entries at a time, use random number sequence to generate password transfer table, that is, put 256 random numbers in a matrix, then sort them, and generate a randomly sorted table according to the initial position. Numbers between 0 and 255. This produces a concrete 256-byte table. Let the random number generator then generate the rest of the numbers in the list, so that each list is different. Next, use the "Shot Gun Technique" technique to generate a decoding table. Basically, if a maps to b, then b must map to a, so b[a[n]]=n (n is a number between 0 and 255). Assign values in a loop so that a 256-byte decoding table corresponds to the previously generated 256-byte encryption table.

Through this method, such a table can be generated. The order of the table is random, so the random number of 256 bytes is generated using a second pseudo-random, and two additional 16-bit passwords are used. Now, there are two conversion tables, the basic encryption and decryption process is as follows: the previous byte ciphertext is the index of this 256-byte table, or, in order to improve the encryption effect, you can use the extra 8-bit value, Even use a checksum or crc algorithm to generate index bytes. Assuming that this table is an array of 256×256, then it satisfies:

encrypt1=a[encrypt0][value]

The variable 'encrypt1' is the encrypted data, and 'encrypt0' is the previous encrypted data (or a function value of the previous encrypted data). Naturally, the first data needs a "seed", which must be remembered.

The pseudo-random sequence generated during encryption is very random and can be designed as any desired sequence. Without detailed information about this random sequence, it is not practical to decrypt the ciphertext. For example: some ASCII code sequence, such as "processing" may be converted into some random meaningless gibberish, each byte depends on the ciphertext of the previous byte, rather than the actual value. For this transformation of any single character, the effective true length of the encrypted data is hidden.

The fourth step is to use the biometric identification method based on bionics and high-dimensional spatial geometry adaptive coverage theory to identify the identity of the examinee.

The present invention uses different methods to calculate the matching degree for the global feature vector and the detail feature vector. For the global feature vector, it is directly mapped to a point in the high-dimensional space, and the matching degree between it and the training sample set is calculated; for the detailed feature, the method of dynamic programming is used for template matching. Then the results of the two are multiplied to obtain the matching degree of a biological feature, and finally the matching degree of multiple biological features is weighted and summed to obtain the overall matching degree, and the recognition conclusion can be drawn after comparing with the threshold.

Taking online signature recognition as an example, firstly, the matching degree of each associated feature vector in the detailed feature is obtained by dynamic programming method with the training sample set, and the matching degree is weighted and summed to obtain the matching degree of the detailed feature. Then the global feature vector is mapped to a point in the high-dimensional space, and the corresponding global feature matching degree is obtained according to the relationship between it and the coverage area of the training sample set. Finally, the details are multiplied by the global feature matching degree to obtain the final online signature overall matching degree. Finally, the matching degree of online signature, face, and pen-holding fingerprint is weighted and summed and compared with the threshold, and finally the identity verification of the examiner is realized.

The fifth step is to dynamically update the training sample set.

After the user passes the identity verification, the new biometric sample information is automatically recorded by the system. When the storage space is limited, the training sample set must dynamically delete some old information while adding new information. This is reasonable in practical applications, so people's signature habits may change over time, and it is meaningful to let the training sample set reflect its changing trend in a dynamically updated manner.

The way to update the training sample set can be considered through the selection of the coverage area of the geometric shape in the high-dimensional space. The present invention realizes the self-adaptive dynamic generation of the coverage area, that is, it is considered that the dynamically updated training sample set should realize the most reasonable dynamic coverage of all samples in the high-dimensional space, and the newly verified biological characteristics are updated to the training sample set, At the same time stale information is filtered out from the coverage area.

Claims (5)

1. a kind of examiner's identification method based on bionics and biometric identification, comprises utilizing collector to collect multiple biometrics, the data of collection is processed, carries out authentication appraisal according to training sample set, it is characterized in that specifically comprising the following steps: (1) Simultaneously collect multiple biometric features of the examinee, and transmit the collected data to the front-end processing device after being encrypted by hardware. The biometric features at least include online signature, pen-holding fingerprint pattern and face image; (2) Preprocessing of the collected data, including filtering and regularizing the data; (3) Feature extraction and feature combination are performed from the preprocessed data, and then the data is soft-encrypted, and then transmitted to the back-end processing equipment through the network, wherein, when combining the extracted biological features, the extracted features are firstly processed Scale division, which is divided into overall information and detailed information, and then further divided according to the correlation between features, the highly correlated features are divided into the same associated feature set, and finally feature fusion is performed to synthesize the associated features into feature vectors; (4) Back-end analysis and processing of data: the back-end processing equipment decrypts the data packets received from the network, and adopts the biometric identification method based on the adaptive coverage theory of high-dimensional spatial geometry to realize the identification of the examinee. The specific identification method Yes, directly map the global feature to a point in the high-dimensional space, and calculate the matching degree between it and the training sample set; for the detailed feature, use the dynamic programming method for template matching, and then multiply the results of the two Obtain the matching degree of a biological feature, and finally obtain the overall matching degree by weighting and summing the matching degrees of multiple biological features, and draw the identification conclusion after comparing with the threshold; (5) Dynamic update of the sample set: add the verified data to the sample set, dynamically generate the corresponding template, and make the new template reflect the trend of the sample change. 2. The examiner's identification method based on bionics and biometric identification according to claim 1, characterized in that: the acquisition method of the step (1) is to first collect the examiner's online signature through the data board, and use human The posture is relatively stable when signing. The camera is used to obtain relatively clear images of the face and signature posture. At the same time, the fingerprint image of the pen holding the pen is collected through the signature pen equipped with a multi-functional sensor at the front. All contact information between pens, including fingerprint texture fragments and relative positions of all contacts. 3. The examiner's identification method based on bionics and biometric identification according to claim 2, characterized in that: gather the characteristics of the pressure of holding the pen while signing, and the method is to set the pressure at each contact of the signature pen The sensor collects the change of the pen-holding pressure during the signing process and records it corresponding to the corresponding area of the pen-holding fingerprint image. 4. The examiner's identification method based on bionics and biometric identification according to claim 1, characterized in that in step (3), the data after feature combination is soft-encrypted using a multi-step encryption algorithm. 5. the examiner's identification method based on bionics and biometric identification according to claim 1, characterized in that the update mode of the training sample set in the step (5) is: the training sample set after the dynamic update is in the high-dimensional space The most reasonable dynamic coverage of all samples should be achieved, that is, the adaptive dynamic generation of the coverage area, updating the newly verified biometrics to the training sample set, and filtering out the old information from the coverage area.

Images