TWI415010B - Face recognition method based on individual blocks of human face - Google Patents

Abstract

A human face recognition method based on individual face regions divides a whole human face into a number of regions, and extracts facial features of said plural regions respectively, and then combines separately recognized matches into a human face recognition result. The method mainly comprises the five steps of dividing the human face, calculating characteristic values among regions, extracting a representative characteristic value among regions, comparing the characteristic value and selecting recognition results. The present method solves problems of being prone to partial shielding of the whole human face, local lighting difference or angle variation and so on, thereby increasing an environmental adoption and accuracy of human face recognition.

Description

Face recognition method based on individual face blocks

The invention relates to a face recognition method based on individual face blocks, in particular to a method of cutting an entire face into a plurality of parts, and respectively using the partial partial block images to perform local feature extraction and comparison of faces. Then, a face recognition method that combines the final identification results is combined.

The general face recognition method mostly uses the image of a whole face to identify it. However, when a whole face image is used as the recognition feature, it is easy to be partially masked by the face, the local light is different or the angle is changed. The eigenvalues of the face of the face produce a large number of changes, which easily lead to the failure of the final identification; later, a face recognition method based on facial features is developed, such as the patent "face recognition method based on facial features" ( Patent Publication No.: 200707310), which proposes to cut the entire facial image into the facial features, including the eyebrows, left eye, right eye, nose, mouth and ears; and then input these facial features into the prior training. In the individual classifiers, according to the distance information and the threshold value, which candidate is selected for the input image, and finally the candidates identified by the facial features are combined to vote for which one of the face databases. ; but this approach will lead to the following problems:

1. The structure of the face is an important basis for identification. However, if the face image is cut too carefully, the structure of each other will be lost.

2. The image of the small area is used as the basis for identification. However, if the algorithm for finding the facial features is not strong, the small deviation in the position will result in a large difference in the image of the facial features.

3. There is no set of standard for the definition of the five-official category, which is described and defined purely by subjective consciousness. However, it is difficult for everyone to agree on the facial features. Therefore, the description of the final identification result is likely to cause users. The use of misunderstandings and confusion arises.

It can be seen that there are still many shortcomings in the above-mentioned methods of use, which is not a good design, but 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 succeeded in researching and developing the face recognition method based on the individual face of the face.

The object of the present invention is to provide a face recognition method based on individual blocks of a face, which is used to improve the influence of partial masking, partial light or angle change of the entire face image by the person cutting the face. The method of face image into several parts, and then extracting features for identification comparison, can improve the environmental adaptability and recognition accuracy of face recognition; face recognition based on individual face of the invention can achieve the above object In the method, the entire face image is used for cutting, and the features are separately extracted for identification, so as to improve the recognition rate; and after the face image is cut into several parts, the LBP method (Reference 1) is used to calculate the face. The eigenvalues of the image block are then selected by Adaboost (Reference 2) to select representative eigenvalues for several partial block images, and used as Adaboost feature selection module; each image block can be utilized for identification. The Adaboost feature selection module selects the feature values and calculates the feature values of each of the relative blocks in the database. The calculated value is determined by the threshold value. This person is compared; each image block to obtain a plurality of possible candidates can be identified The candidate for the result is finally selected by the decision module of Borda Count (Reference 3) to select the final identification result. The candidate selected by the Borda Count decision module is not necessarily the one with the most votes, and will pass different weight values. Taking into account the overall trend, the leading trend is to obtain the most identification result after the weight calculation. In addition, there is a threshold in the Borda Count decision module. When considering the weight and ranking, the score is calculated. The highest candidate is accepted if the score is above the threshold; if it is below the threshold, the person is considered illegal.

Please refer to FIG. 1 , which is a system architecture diagram of a face recognition method based on individual face blocks of the present invention. As can be seen from the figure, the system mainly includes: an image input module 1 and a face image. The cutting module 2 is connected to the face image input module 2, and the face image cutting module 2 is connected with an image input module 1 and an LBP feature module 3. The face image input by the image input module 1 is cut into a plurality of face image images in a horizontal, vertical or horizontal manner; an LBP feature module 3 is connected to the face image cutting module 2 The Adaboost feature selection module 4 is connected. The LBP feature module 3 uses the LBP method to find the feature value of the cut face image block; an Adaboost feature selection module 4, the system and the LBP feature The module 3 and the comparison module 5 are connected, and the Adaboost feature selection module 4 selects the most representative feature value for each image block; a comparison module 5, and the Adaboost feature selection module 4, face block image database 6 And the Borda Count decision module 7 is connected, the comparison module 5 can use the Adaboost feature selection module 4 to select the feature values, respectively, and the feature values of the relative blocks of each person in the face block image database 6 Calculating, the calculated value uses the threshold value to determine whether the person is a comparison; a face block image database 6 is connected to the comparison module 5, and the face block image database 6 includes There is a feature value of each person relative block; a Borda Count decision module 7 is connected with the comparison module 5, and the Borda Count decision module 7 is responsible for selecting the final identification result.

Please refer to FIG. 2 , which is a flowchart of a face recognition method for a face recognition method based on individual face blocks of the present invention. The steps are as follows: 1. When performing recognition, a face Image input 101, through the face image cutting module first finds the face position (for face position positioning), and cuts the face image into three equal parts (three image blocks) 102; 2. will cut this In the halving area, the eigenvalues 103 of the image blocks (three image blocks) of the human face can be calculated through the LBP feature module; 3. The image feature values obtained from the face images of the three image blocks are respectively input. In the Adaboost feature selection module 12, and then through the previously trained selector, each image block will select the most representative feature value 104; 4. Input these feature values into the comparison module to identify the image block. The eigenvalue and the score of the eigenvalue of each person in the database relative to the block, the absolute value of the two-two phase subtraction is used, and the value is used to determine whether the person is the comparison 105; 5. Through the comparison module After comparison, each image block will generate several Candidates can Identification results, these candidates are ranked from the highest score to the lowest score, and finally the Borda Count decision module is used to select the final result (using the Borda Count decision module combined with the identification results of each image block to select the final identification result. ) 106.

Before performing face recognition, the Adaboost feature selection module must be trained in advance as the feature selector for identification. The Adaboost feature selection module is trained to prepare two types of data. Each person has several desires. The image of the training, the first type is to subtract the absolute value of the image of the same person, and the second type is to subtract the absolute value of the image of the different people; then the Adaboost feature selection module is initialized, The weights of the two types of data to be distinguished are set to be the same, and the best-performing weak classifiers are selected from a plurality of weak classifiers (Reference 2) per round in a recursive manner, according to which The classifier's answer updates the weight of each data. The way to update is to reduce the weight of the data in the correct answer, and the weight of the wrong data is increased. The purpose is to make the weak classifier selected in the next round to rectify the wrong answer in this round. Finally, combining these weak classifiers into a strong classifier (Reference 2) becomes the Adaboost feature selection module. According to this training method, each image area of the face image is separately selected. After the block is processed, the Adaboost feature selection module can be trained.

Please refer to FIG. 3 , which is a schematic diagram of a face image cutting module of a face recognition method based on individual face blocks of the present invention. It can be seen from the figure that the face image cutting module has first found a face. The area, so the cut three equals are just the three important parts, the eyes and the eyebrows, the nose and the mouth.

Please refer to FIG. 4 , which is a schematic diagram of the implementation of the Borda Count decision module of the face recognition method based on the individual face of the present invention. In the process of the face recognition method, the Borda Count decision module will rank. The last one gives S ₀ points, the countdown to two places gives S ₁ points, the analogy to the first place gives S _N points, and S ₀ <S ₁ <...<S _N ; as shown in Figure 4, A gets The score is 51*3+0*(5+23+21)=153, B gets a score of 205, C gets a score of 151, D gets a score of 91, and the highest score is B; except for considering the ranking In addition, the Borda Count decision module also considers the weight value, so the final result selected is not necessarily the candidate with the most votes. The different weight values will take into account the overall trend; otherwise, the trend of weights may not be considered. According to the number of votes, ABCD is four votes, and the final result may be one of them. However, if weights are considered, in the three cases of weight = 5, weight = 23, and weight = 21, the ranking of B is higher than A, so the final calculation At the time of the score, B’s score wins, and the leader is calculated by weight. After that, the most score is the final result; and the Borda Count decision module has a threshold value. When the candidate with the highest weight and ranking score is considered, the score is higher than the threshold. The result is accepted; if it is below this threshold, the person is considered to be an illegal person.

The face recognition method based on the individual face of the face provided by the invention has the following advantages when compared with other conventional techniques:

1. The invention can overcome the influence of partial masking of the face, local light or angle change to improve the recognition rate, thereby enhancing the environmental adaptability of the face recognition.

2. The invention can reduce the time of face training. When the database is added with new users, it is not necessary to retrain the Adaboost feature selection module to increase the overall efficiency of the system.

3. The invention can effectively cut the face area. Because it is not cutting the small area of the facial features, but only cutting the face image equally, the error of the cutting of the facial features is reduced, so as to increase the accuracy of cutting and maintain The structure between the five senses to achieve the purpose of improving the recognition rate.

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 be submitted in accordance with the law in accordance with the statutory invention patents that fully meet the novelty and progressiveness, and you are requested to approve this article. Invention patent application, in order to invent invention, to the sense of virtue.

1‧‧‧Image input module

2‧‧‧Face image cutting module

3‧‧‧LBP feature module

4‧‧‧Adaboost feature selection module

5‧‧‧Comparative Module

6‧‧‧Face block image database

7‧‧‧Borda Count Decision Module

1 is a system architecture diagram of a face recognition method based on individual face blocks of the present invention; FIG. 2 is a flowchart of a face recognition method for a face recognition method based on individual face blocks of the present invention. FIG. 3 is a schematic diagram of a face image cutting module based on a face recognition method based on an individual face of the present invention; and FIG. 4 is a face recognition method based on an individual face of the present invention. Schematic diagram of the implementation of the Borda Count decision module.

Claims (5)

A face recognition method based on individual face blocks, the steps of which are: (1) inputting a face image and cutting the face image by using a face image cutting module to perform face position positioning. (2) using the LBP feature module to calculate the feature values of each image block; (3) using the Adaboost feature selection module to select representative features of each image block through the previously trained selectors. (4) using the comparison module to calculate the feature value of the identified image block and the score of the feature value of each person in the database, and then use the threshold value to determine whether the person is compared; (5) utilizing The Borda Count decision module combines the identification results of each image block to select the final identification result. The face recognition method based on the individual face of the face according to the first aspect of the patent application, wherein the face image cutting module is capable of cutting into a plurality of personal face block images in a horizontal manner. The face recognition method based on the individual face of the face according to the first aspect of the patent application, wherein the face image cutting module is capable of cutting into a plurality of personal face block images in a vertical manner. The face recognition method based on the individual face of the face according to the first aspect of the patent application, wherein the face image cutting module is capable of cutting into a plurality of personal face block images in a horizontal and vertical combination manner. Face recognition based on individual face blocks as described in item 1 of the patent application scope The method, wherein the Borda Count decision module has a threshold value, and when the candidate with the highest weight and the highest score is calculated, if the score is higher than the threshold, the result is accepted. If the threshold is lower than the threshold, The person is considered to be an illegal person.