TWI424359B - Two - stage Face Recognition System and Method - Google Patents

Description

Two-stage face recognition system and method

The invention relates to a two-stage face recognition system and method, in particular to a method for utilizing two stages of classification and verification, which accelerates the identification time by using a computationally intensive support vector machine (SVM) classification method, and then utilizes more accurate The feature comparison method is verified to improve the face recognition accuracy.

Face recognition is a research technology emerging in recent years. Its non-contact and high-convenience advantages are widely valued by researchers and industry. They are expected to have excellent management in public security or access control. Performance; a successful face recognition system can be applied to identity identification at the entrance and exit of the building, identification of suspects, and even to the identification of entry and exit identification for maintaining aviation safety. In the current face recognition system, the identification program is for the person standing in the position to prepare for identification, and the face or side image of the face is synchronized by the image capturing module, and is extracted through the feature and all facial features of the personnel database. Compare, and then find the lowest-discussed personnel database members, and finally determine the image to be recognized as a registered/unregistered person by a threshold value. This practice has been exposed to the patent "applying in a setting environment." Method and apparatus for face detection and recognition" (Patent Publication No.: 200709075).

In addition, as in the paper [1], the face region is divided into several small regions, and each pixel of the face region is image-encoded by the LBP (Local Binary Pattern) method, and the central pixel and the eight-neighbor pixel value are calculated. Correlation, in order to generate a two-digit code, the facial features are obtained by the histogram (histogram) counting of the small-area binary code, and the difference between the test image and the training data is calculated using Chi-square distance [2]. Degree is used as the identification classifier to produce the identification result. In the paper [3], the MBLBP (Multi-scale Block Local Binary Patterns) method is used for facial feature calculation, and then the AdaBoost module [4] is used as the identification classifier. All of the above identification systems and methods have a common point. When a test image input is to be recognized, it must be compared one-to-one with the facial features of all personnel in the personnel database (training data), thereby obtaining the lowest degree of difference. The identification of the personnel results, but with the expansion of the personnel database, will result in increased computational complexity and data comparison time, thereby reducing the practicality of face recognition products.

It can be seen that there are still many shortcomings in the above-mentioned methods of use. It is not a good design and needs to be improved.

In view of the shortcomings derived from the above-mentioned conventional methods, the inventor of the present invention has improved and innovated, and after years of painstaking research, he finally successfully developed a two-stage face recognition system and method.

The object of the present invention is to provide a two-stage face recognition system and method, which mainly utilizes a two-stage method of classification and verification to achieve an accelerated recognition time and an improved face recognition accuracy.

The two-stage face recognition system and method for achieving the above object of the invention is to improve the current face recognition method, which requires a large computational complexity and data comparison time, and adds a verification procedure to improve the accuracy of face recognition. Sex. Each person needs to register before identification, and take a series of face images with multiple face angles. After the program software selects the image with suitable angle, the face feature data in the face image is extracted and stored in the personnel database. Through the support vector machine (SVM) algorithm [5], the multi-class training of the facial features of the personnel database is used to generate support vector machine (SVM) classification modules, each category consists of a person's face. The angle feature data is composed; when the image input to be recognized is to be recognized, the facial feature data is calculated, and the facial feature data is classified by the support vector machine (SVM) classification module, and the most similar to the registered The classification result of a member of the personnel database; the face verification is performed by the feature comparison method, and the facial feature data to be identified and the facial feature data of the registered person are compared one-to-many, and the verification method is verified. Use Chi-square distance[2] to calculate the difference between the facial features. If the difference between the facial features is lower than a threshold, the image to be identified is considered to be the same as the registered person. One person outputs the name of the person; if the difference between the facial features of the two is higher than a threshold, it is considered that the image to be recognized is not the same person as the registered person, and the identification result of the unregistered person is output.

Referring to FIG. 1 , an architecture diagram of a two-stage face recognition system and method of the present invention includes: a face capture module 10 for capturing an image of a face, and then a feature extraction module 11 The feature extraction module 11 calculates a facial feature by using a feature extraction method to obtain a facial feature data 12; wherein the feature extraction method of the feature extraction module 11 can be a feature value calculation ( Local Binary Patterns (LBP) image coding method, or Multi-scale Block Local Binary Patterns (MBLBP) image coding method, plus AdaBoost learning method; A vector machine (SVM) classification module 13 calculates the above facial feature data 12 and obtains a member of the most similar registered personnel database, using a support vector machine (SVM) algorithm [5], for personnel The database 14 is a class classification module obtained by multi-category training, each category is composed of a plurality of facial angle feature data of a person; a personnel database 14 stores the facial feature data of all personnel; The facial feature data of each person includes characteristic data of multiple and multi-face angle images, and the suitable image can be selected by software or manual; a database extraction module 15 is adopted by the face extraction mode. The group 10 and the facial feature data 12 of the feature extraction module 11 are combined with the classification result of the support vector machine (SVM) classification module 13, and the specific member facial feature data 16 is selected from the registered personnel database 14 The member face exogenous data of the classification result is selected for use as the verification comparison; a facial feature verification module 17 is used to make the facial feature data 12 and the specific member facial feature data 16 For the verification comparison, the verification comparison method uses the Chi-square distance to calculate the difference of the facial features. If the difference is lower than the threshold, the image to be recognized and the member are considered as The same person, and output the member name; if the difference between the facial features of the latter two is higher than a threshold, it is considered that the image to be recognized is not the same person as the member, and the non-registered person is output; wherein, the threshold value Via calculation The Chi-square distance between the facial features of the personnel database 14 includes Chi-square distance for calculating the facial features of the person and the person and the Chi-square distance of the facial features of the person and the non-self, and then according to the two types of Chi-square distance. The square distance is determined by the average.

Among them, the computer can be read by a computer with a built-in program, and the two-stage face recognition system and method described in the present case can be completed when the computer is loaded and executed.

In addition, in order to test the two-stage face recognition system and method of this case, the experiment was carried out with the face database taken by the laboratory. The face database consisted of 100 personnel, and each person took 60 consecutive photos. The face image of the face angle has an image resolution of 320×240. The suitable 30 images are filtered into two parts, 15 pieces of training data and 15 pieces of test images to be identified. The implementation method is to randomly select 80 persons from 100 people to register, that is, store the facial features of the 80 personnel training materials in the personnel database, and use the support vector machine (SVM) algorithm to the personnel. The database is organized into 80 categories to generate a staff classification module. Each category consists of a plurality of facial angle features of a person. Through the steps of the two-stage face recognition method, 100 test personnel are to be identified. If a person is considered to be a registered person in more than 5 of the 15 test images, the registered person's name is output; otherwise, if a person does not exceed 5 in 15 test images, it is considered to be a certain Registered personnel, the results of non-registered personnel are output; the two-stage process of classification and verification can greatly reduce the recognition time, compared to the facial features of all personnel in the personnel database for one-to-one comparison [3], The recognition time is reduced from 1.206 seconds per person to 0.548 seconds per person; the verification mechanism can improve the accuracy of face recognition. Compared with the identification method using only support vector machine (SVM), the recognition rate is 98.75%. However, the staff rejection rate has increased from 88% to 98%.

Please refer to FIG. 2, which is a flow chart of the steps of the two-stage face recognition system and method of the present invention, including: Step 1: The face capture module captures the face image 201 to be recognized; Step 2: utilizes the feature extraction mode The group extracts the facial feature data 202; step 3: the facial feature data is calculated by the support vector machine (SVM) classification module, and a member 203 of the most similar registered personnel database is obtained; step 4: the database extraction module Selecting the facial feature data of the specific member in the personnel database, that is, selecting the facial feature data 204 of the classification result; Step 5: The facial feature verification module compares the facial features of the personnel database with the facial features of the specific member Compliance with step 205; Step 6: If yes, the verification succeeds 206, and the member name 207 is output; Step 7: If not, the verification fails 280, and the message 209 of the non-registered person is output.

The two-stage face recognition system and method provided by the invention have the following advantages when compared with other conventional technologies:

1. The invention can greatly reduce the computational complexity and speed up the identification time, so that it can be applied to the face recognition system of the larger personnel database.

2. The invention can improve the face recognition accuracy, and can be applied to the face recognition system to ensure the safety of home and access control management.

The detailed description of the preferred embodiments of the present invention is intended to be limited to the scope of the invention, and is not intended to limit the scope of the invention. The patent scope of this case.

To sum up, this case is not only innovative in terms of technical thinking, but also able to enhance the above-mentioned multiple functions compared with conventional articles. It should fully comply with the statutory invention patent requirements of novelty and progressiveness, and apply in accordance with the law. I urge you to approve this article. Invention patent application, in order to invent invention, to the sense of virtue.

10. . . Face capture module

11. . . Feature extraction module

12. . . Facial feature data

13. . . Support vector machine (SVM) classification module

14. . . Personnel database

15. . . Database extraction module

16. . . Specific member facial feature data

17. . . Facial feature verification module

1 is an architectural diagram of a two-stage face recognition system and method of the present invention;

2 is a flow chart of the steps of the two-stage face recognition system and method of the present invention.

10. . . Face capture module

11. . . Feature extraction module

12. . . Facial feature data

13. . . Support vector machine (SVM) classification module

14. . . Personnel database

15. . . Database extraction module

16. . . Specific member facial feature data

17. . . Facial feature verification module

Claims (7)

A two-stage face recognition system includes: a face capture module for capturing an image of a face, and then a feature extraction module for calculating a facial feature; a feature extraction module, the feature of which is used The extraction method calculates the facial features and obtains the facial feature data; a support vector machine (SVM) classification module calculates the facial feature data and obtains a member of the most similar registered personnel database; The library stores the facial feature data of all the personnel; a database extraction module selects the facial feature data of the specific member from the registered personnel database, that is, selects the facial feature data of the member of the classification result. Used as a verification comparison; a facial feature verification module is used to perform one-to-many verification comparison between facial feature data and specific member facial feature data. For example, the two-stage face recognition system described in claim 1 is characterized in that the feature extraction module calculates a facial feature by using a feature value calculation (LBP) image coding method. For example, the two-stage face recognition system described in claim 1 is characterized in that the feature extraction module calculates a face by using a multi-resolution block local binary calculation (MBLBP) image coding method and an AdaBoost learning method. feature. The two-stage face recognition system according to claim 1, wherein the facial feature verification module calculates a difference in facial features using a Chi-square distance. A two-stage face recognition method comprises the following steps: Step 1: The face capture module captures the face image to be recognized; Step 2: Extracts the face feature data by using the feature extraction module; Step 3: The facial feature data is calculated by the support vector machine (SVM) classification module to obtain a member of the most similar registered personnel database; step 4: the database extraction module selects the face of the specific member in the personnel database. The feature data is the member facial feature data of the selected classification result; Step 5: The facial feature verification module compares the facial features of the personnel database with the facial features of the specific member; Step 6: If it is met, the verification is performed. Success and output member name; Step 7: If not, the verification fails and the message of the non-registered person is output. A computer readable medium in which a program is stored, and when the computer is loaded with a program and executed, the system as described in claim 1 can be completed. A computer readable medium having a built-in program, which can be completed as described in claim 5 after the computer is loaded and executed.