Computer equipment applied to face recognition

Abstract

The invention discloses computer equipment applied to face recognition, and the computer equipment comprises a memory, a processor and a computer program which is stored in the memory and runs on the processor, so that the face angle span of multi-angle face recognition is enriched; the algorithm is improved on the basis of the original GMRF, parameters are selected according to experiments, multi-angle face images can be classified by using a computer program, the image classification precision is improved, the method is more suitable for the real environment, and a head posture database is utilized to realize the multi-angle face image classification. The aim that each gesture has two pieces is to be able to train and test algorithms on known and unknown faces. The face position on each image is marked in an independent text file, and under the condition that experimental data are the same, the information amount of the GMRF features on the multi-angle face images is increased, the coverage of the face angle is enriched, and the accuracy is improved. Multi-angle face images can be effectively classified, and the classification efficiency is improved.

Description

[0001] This application is a divisional application with application number 202310525186.1, application date May 10, 2023, and the invention name at the time of application is “A method for extracting multi-angle facial features”. Technical Field

[0002] The present invention relates to the technical field of face recognition, and in particular to a computer device applied to face recognition. Background Art

[0003] In recent years, with the expansion of terminals, the maturity of internet technology, and the widespread use of electronic devices, video surveillance has not only brought convenience to people but also provided strong protection for the safety of life and property. Due to the perspective of the face in surveillance, a person can appear from multiple angles, and how to quickly identify the face based on local angles is a problem that needs to be solved.

[0004] In recent years, with the advent of deep learning theory, many new face recognition methods have emerged. Gaussian Markov random field models (GMRFs) have shown promising results in extracting texture features from images, but further research on GMRF models is lacking, and they lack selective treatment of the characteristics of face recognition, for example, ignoring facial texture information. The order of the GMRF must be selected in real time for different problems. Furthermore, when selecting spatial pixels within a GMRF, traditional methods rely on linear calculations, making comprehensive analysis of complex textures difficult.

[0005] The existing GMRF feature extraction method has shortcomings in the description of multi-angle face image features, such as too few local area divisions, missing texture information, and missing spatial information of facial texture structure, which leads to poor multi-angle face image classification effect. Summary of the Invention

[0006] The purpose of the present invention is to provide a computer device applied to face recognition to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a multi-angle facial feature extraction method, comprising the following steps:

[0008] Step 1: Collect and import the module, take and store color face images;

[0009] Step 2: Grayscale processing module, grayscale processing and saving the color face image, performing face segmentation and face Fourier transform on the grayscale face image to form the frequency domain information of the new processed face image in different regions;

[0010] Step 3: Generate a module that pre-processes the frequency domain information of the face images in different regions in step 1 into blocks to generate different feature image blocks;

[0011] Step 4: In the calculation module, different feature image blocks enter the GMRF model to extract feature points to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing;

[0012] Step 5: The merging module combines the GMRF features extracted in step 3, labels the face categories, and packages them together with the feature vectors in step 2;

[0013] Step 6: Output module, put the feature vectors and labels packaged in step 4 into the SVM classifier for training and recognition, and obtain the classification results by comparing with other multi-angle algorithms.

[0014] Furthermore, in step 2, the step of extracting frequency domain information after graying the color face image is as follows: graying the face image and converting it into a double-precision two-dimensional array, and then performing Fourier transform on the converted two-dimensional array.

[0015] Furthermore, in step three, the step of dividing the characteristic image into blocks is as follows: reading the frequency domain image features after Fourier transformation, and respectively calculating the length and width of the characteristic image;

[0016] The location of the block points of each method is calculated according to the size of the feature image block. After the calculation, the processed feature image is taken and saved according to different blocks.

[0017] Furthermore, in the step 4, the step of forming the GMRF features after block processing is as follows: first reading the different feature image blocks saved in the step 3;

[0018] Then, the feature points of different feature image blocks are extracted using GMRF features to form feature point vectors with calculations.

[0019] Furthermore, in the step 4, after the feature vectors of each feature image block are obtained, they are concatenated together to form a feature vector of the entire feature map and packaged together with the label of the feature map to prepare for subsequent training.

[0020] Furthermore, in step six, the feature vector and SVM classification algorithm are used to classify multi-angle face images, and the classification accuracy is compared by utilizing the influence of GMRF models of different orders on the classification. The steps for determining whether the classification result is correct are as follows:

[0021] Use SVM multi-classification algorithm to classify multi-angle face images based on feature vectors;

[0022] The average accuracy results are then used for evaluation.

[0023] A device for implementing a multi-angle facial feature extraction method, the device comprising:

[0024] Collection and import module, used to capture or import color face images;

[0025] Grayscale processing module, used for grayscale processing module, grayscale processing and saving of color face images, face segmentation and face Fourier transform of the grayscale face images, to form frequency domain information of new processed face images of different regions;

[0026] A generation module is used to pre-process the frequency domain information of the face image in different regions in step 1 into blocks to generate different feature image blocks;

[0027] The calculation module is used to extract feature points from different feature image blocks into the GMRF model to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing;

[0028] The merging module is used to combine the GMRF features extracted in step 3, label the face categories, and package them together with the feature vector in step 2;

[0029] The output module is used to input the feature vectors and labels packaged in step 4 into the SVM classifier for training and recognition, and obtain the classification results by comparing with other multi-angle algorithms.

[0030] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the multi-angle facial feature extraction method are implemented.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-angle facial feature extraction method.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention enriches the facial angle span of multi-angle face recognition, improves the algorithm based on the original GMRF and selects parameters according to experiments. Computer programs can be used to classify multi-angle facial images, improve the accuracy of image classification, reduce the workload, and be more suitable for real-world environments.

[0034] The face database chosen for this invention is the head pose database, a benchmark of 2,790 monocular face images of 15 people, with translation and tilt angles ranging from -90 degrees to +90 degrees. For each person, two series of 93 images are available. The purpose of having two images for each pose is to be able to train and test the algorithm on known and unknown faces. The people in the database have different skin colors and whether they wear glasses. The background is natural, neutral, and uncluttered to focus on facial manipulation. The location of the face on each image is marked in a separate text file.

[0035] This method focuses on feature extraction from facial images at different angles. Compared to other algorithms, it offers robustness and excellent classification results. The van der Lugt correlator combined with zero-mean normalized cross-correlation achieves face recognition from different angles, achieving an overall recognition rate of 75.38%.

[0036] The facial region was extracted from the original multi-angle face image and resized to 128*128 pixels. The extracted face image was then grayscaled and divided into blocks, which were then fed into the GMRF model. Feature extraction and splicing were then performed for classification and recognition. The classification performance was 70.29%, which was 5.09% lower than the previous method, not only failing to improve. This may be due to the asymmetry of the pixels in the image space during image resizing, which negatively impacted the GMRF's feature point selection.

[0037] After graying and blocking the face image, the second-order GMRF is used for feature extraction and classification. The classification effect is 77.39%, which is 2.01% higher than the above method. Since the low-order GMRF model pays too much attention to the description of details, it ignores the relationship between the whole. Therefore, the present invention improves the GMRF order based on the second order. Experiments were conducted on the 2nd, 4th, 5th, 8th and 14th order Gauss-Markov models respectively. The highest recognition rate reached 94.78%, which is 17.39% higher than the 2nd order. The experiment proves that the high-order GMRF model has a robust description capability for the angle changes of the face.

[0038] Considering the impact of Fourier transform on the selection of facial feature points, the real and imaginary parts, as well as the modulus and angle after Fourier transform were extracted and calculated. GMRF feature extraction was then performed, and finally recognition and classification were performed. The best result was 99.57%.

[0039] In summary, using the same experimental data, this invention increases the information content of GMRF features in multi-angle facial images, enriches the coverage of facial angles, and improves accuracy. It can effectively classify multi-angle facial images, improve classification efficiency, and reduce the workload of manual classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 A schematic diagram of the GMRF feature point extraction method according to the present invention;

[0042] Figure 3 The present invention relates to a schematic diagram of extracting a face area;

[0043] Figure 4 This is a schematic diagram of features obtained by Fourier transforming a multi-angle face image according to the present invention;

[0044] Figure 5 A schematic diagram of a multi-blocking method for multi-angle facial images according to the present invention;

[0045] Figure 6a The comparison of the face image extraction and reconstruction method of the present invention on the classification effect of multi-angle face images at different GMRF orders;

[0046] Figure 6b This is a comparison of the classification effects of multi-angle face images using different GMRF orders when directly extracting face information in the present invention;

[0047] Figure 6c The present invention compares the classification effects of multi-angle face images at different GMRF orders after Fourier transformation.

[0048] Figure 7 A partial display of the multi-angle face database used in this invention;

[0049] Figure 8 A schematic diagram of the composition of a multi-angle facial feature extraction device provided by an embodiment of the present invention;

[0050] Figure 9 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] See also Figure 1-9 The present invention provides a technical solution: a multi-angle face feature extraction method, comprising the following steps:

[0053] Step 1: Collect and import modules, take and store color face images

[0054] Step 2: Grayscale the color face image and save it, perform face segmentation and face Fourier transform on the grayscale face image to form frequency domain information of new processed face images in different regions;

[0055] Step 3: Preprocess the frequency domain information of the face images in different regions in step 1 into blocks to generate feature image blocks;

[0056] Step 4: Different feature image blocks enter the GMRF model to extract feature points to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing;

[0057] Step 5: Concatenate the GMRF features extracted in step 3, label the face categories, and package them together with the feature vector in step 2;

[0058] Step 6: The feature vectors and labels packaged in step 4 are put into the SVM classifier for training and recognition, and the classification results are obtained by comparing with other multi-angle algorithms.

[0059] The face database selected by the present invention is the Head Pose Image Database (HPID), which is a benchmark of 2,790 monocular face images of 15 people with translation and tilt angles ranging from -90 degrees to +90 degrees. For each person, two series of 93 images (93 different poses) are available. The purpose of having two images for each pose is to be able to train and test the algorithm on known and unknown faces. The people in the database have different skin colors, whether they wear glasses or not. The background is natural, neutral and tidy to focus on facial operations. The location of the face on each image is marked in a separate text file.

[0060] The Vander Lugt correlator (VLC) and zero-mean normalized cross-correlation (ZNCC) achieve face recognition at different angles, with an overall recognition rate of 75.38%.

[0061] In step 2, the step of extracting frequency domain information after graying the color face image is as follows: graying the face image and converting it into a double-precision two-dimensional array, and then performing a Fourier transform on the converted two-dimensional array. The Fourier transform F(u, v) of the two-dimensional continuous function f(x, y) is defined as:

[0062]

[0063] If f(x, y) is a real function, its Fourier transform is symmetric, that is:

[0064] F(u,v)=F(-u,-v)

[0065] The frequency spectrum of the Fourier transform is symmetrical |F(u,v)|=|F(-u,-v)|

[0066] Given F(u, v), f(x, y) can be obtained by inverse Fourier transform:

[0067]

[0068] Two-dimensional polar coordinate representation of the Fourier transform:

[0069]

[0070] The amplitude or frequency spectrum is:

[0071]

[0072] R(u, v) and I(u, v) are the real and imaginary parts of F(u, v) respectively. The phase angle or phase spectrum is:

[0073]

[0074] The power spectrum is:

[0075] P(u, v)=|F(u, v)| 2 =R(u,v) 2 +I(u,v) 2

[0076] Origin transformation of F(u, v):

[0077]

[0078] Multiply f(x, y) by (-1)x+y to transform the origin of F(u, v) to (M / 2, N / 2) in rate coordinates, which is the center of the M×N area u=0, 1, 2, ..., M-1, v=0, 1, 2, ..., N-1.

[0079] The correspondence between frequency components and image appearance can be exploited. Some enhancement tasks that are difficult to express in the spatial domain become very common in the frequency domain. Furthermore, filtering is more intuitive in the frequency domain, which can explain certain properties of spatial filtering. For multi-angle facial images, the angular transformations have completely different properties when processed in the spatial and frequency domains.

[0080] In the step 3, the step of dividing the characteristic image into blocks is as follows: reading the frequency domain image features after Fourier transformation, and respectively calculating the length and width of the characteristic image;

[0081] The location of the block points of each method is calculated according to the size of the block. After the calculation, the processed feature image is taken and saved according to different blocks.

[0082] In the step 4, the step of forming the Gaussian Markov field GMRF feature after block processing is as follows: first read the block feature image block saved in the step 3;

[0083] Then extract the feature points of GMRF features of different orders to obtain feature point vectors with calculations. The specific steps of the feature point extraction method are as follows:

[0084] Let S be a point set on an M×M network, S = {(m, n), 1≤m, n≤M}, and assume that the given texture [y(s), s∈S, S = {(m, n), 1≤m, n≤M}] is a zero-mean Gaussian random process. The GMRF model can be represented by a linear equation containing multiple unknown parameters. The specific formula is as follows:

[0085]

[0086] Among them, N S represents the GMRF neighborhood of point S, θ r represents the coefficient, e(s) is a Gaussian noise sequence with a mean of zero. Because the pixel neighborhood is symmetrical, θ r =θ-r, the above formula can be written as:

[0087]

[0088] Where y1(s+r) is a point in the closed annular region S. Applying formula (2-27) to each point in region S yields M. 2 Equations related to {e(s)} and {y(s)}:

[0089]

[0090] …

[0091]

[0092] …

[0093]

[0094] …

[0095]

[0096] The system of equations consisting of all y1(s+r) can be expressed in matrix form as follows:

[0097] y=Q T θ+e

[0098] The above formula is the linear model of Gaussian Markov random field, Q T is the matrix of all y1(s+r), and θ is the eigenvector to be estimated of the model. Take the second-order GMRF model as an example:

[0099]

[0100] Neighborhood: N = {(1, 1), (1, 0), (1, -1), (0, 1), (0, -1), (-1, 1), (-1, 0), (-1, -1)}

[0101] Mean and variance: e(m,n)~N( 0 ,σ 2 );

[0102] For each pixel, we use the covariance matrix μ, σ defined in a window W and the parameters {θ(t, s), (t, s) ∈ N}, through least squares estimation (LSE):

[0103]

[0104] N w Represents the number of pixels in the window W. Due to the symmetry of the correlation function, we can estimate the four θ parameters. We can obtain a feature space {μ, σ, θ(t, s), (t, s)∈N}.

[0105] The following estimated solution formula can be obtained.

[0106]

[0107] In the formula It is an asymptotically consistent estimate of the parameters of the Gaussian Markov random field model, and gives the squared error of the GMRF model parameter estimate

[0108] G matrix algorithm model of each order GMRF model (for a sliding window, each non-edge pixel can be calculated as a column vector, which is a column of the G matrix. Therefore, the number of rows of the G matrix is ​​the number of non-edge pixels, and the number of columns is the number of estimated parameters (determined by the order of the GMRF model):

[0109] (1) Second-order model

[0110]

[0111] {r1, r2, r3, r4} = {(0, 1), (1, 0), (1, 1), (1, -1)}, θ is a 4-dimensional vector θ = (θ1, θ2, θ3, θ4);

[0112] (2) Fourth-order model

[0113]

[0114] {r1, r2, r3, r4} = {(0, 1), (1, 0), (1, 1), (1, -1)}, θ is a 10-dimensional vector θ = (θ1, θ2, θ3, θ4, θ5, θ6, θ7, θ8, θ9, θ 10 ), is a 10-dimensional vector;

[0115] (3) Fifth-order model

[0116]

[0117] {r1, r2, r3, r4} = {(0, 1), (1, 0), (1, 1), (1, -1)}, θ is a 10-dimensional vector θ = (θ1, θ2, θ3, θ4, θ5, θ6, θ7, θ8, θ9, θ 10 ,θ 11 ,θ 12 ), is a 12-dimensional vector;

[0118] (4) Eighth-order model

[0119]

[0120] {r1, r2, r3, r4} = {(0, 1), (1, 0), (1, 1), (1, -1)}, θ is a 10-dimensional vector θ = (θ1, θ2, θ3, θ4, ..., θ 19 ,θ 20 ,θ 21 ,θ 22 ), is a 22-dimensional vector;

[0121] (5) Fourteenth-order model

[0122]

[0123] {r1, r2, r3, r4} = {(0, 1), (1, 0), (1, 1), (1, -1)}, θ is a 10-dimensional vector;

[0124] θ=(θ1, θ2, θ3, θ4,..., θ 37 ,θ 38 ,θ 39 ,θ 40), is a 40-dimensional vector.

[0125] In the fourth step, after the feature vectors of each feature image block are obtained, they are concatenated together to form a feature vector of the entire feature map and packaged together with the label of the feature map to prepare for subsequent training.

[0126] In step 6, the feature vector and SVM classification algorithm are used to classify multi-angle face images, and the classification accuracy is compared by using the influence of GMRF models of different orders on the classification. The steps to determine whether the classification result is correct are as follows:

[0127] Use SVM multi-classification algorithm to classify multi-angle face images based on feature vectors;

[0128] Then use the average precision result (MAP) for evaluation, where:

[0129]

[0130] It should be noted that the present invention is based on the general face recognition system framework and modifies the previous facial feature extraction process in four parts to address the characteristics of multi-angle facial images. The four parts are: feature preprocessing of multi-angle faces, a multi-angle facial image segmentation algorithm, GMRF extraction of facial image feature points, and GMRF calculation of feature points to form feature vectors. The method of the present invention uses a set computer execution program to classify multi-angle facial images, with the aim of expanding the angles of facial image classification and improving the efficiency of facial recognition, saving human resources.

[0131] In the model representation of different orders of GMRF that only changes the feature point extraction, the classification effect is proportional to the increase of the order of Gaussian Markov random field. When the order is 14, the expected effect is the best, reaching 94.78%, which is 3.33% and 0.29% higher than the 5th and 8th order GMRF models respectively.

[0132] Considering the superior performance of the Fourier transform in images, this paper also pre-processed multi-angle facial images using the Fourier transform. Classification experiments were conducted on the real and imaginary parts of the Fourier transform, the spectrum, the phase angle, and their combinations. Using the same GMRF model, the real part alone achieved the highest recognition performance, reaching 99.57%.

[0133] In addition, the classification effects of other classifiers were compared, and the results showed that this feature performed best under the SVM multi-classifier among the three.

[0134] A device for implementing a multi-angle facial feature extraction method, the device comprising:

[0135] Collection and import module, used to capture or import color face images;

[0136] Grayscale processing module, used to grayscale the color face image and save it, perform face segmentation and face Fourier transform on the grayscale face image, and form frequency domain information of the new processed face image in different areas;

[0137] A generation module is used to pre-process the frequency domain information of the face image in different regions in step 1 into blocks to generate different feature image blocks;

[0138] The calculation module is used to extract feature points from different feature image blocks into the GMRF model to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing;

[0139] The merging module is used to combine the GMRF features extracted in step 3, label the face categories, and package them together with the feature vector in step 2;

[0140] The output module is used to input the feature vectors and labels packaged in step 4 into the SVM classifier for training and recognition, and obtain the classification results by comparing with other multi-angle algorithms.

[0141] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the multi-angle facial feature extraction method are implemented.

[0142] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-angle facial feature extraction method.

[0143] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, physical sub-libraries and tables or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0144] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A computer device for face recognition, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, a method for extracting facial features is implemented. The specific steps include: Step 1: Collect and import the module, take and store color face images; Step 2: Grayscale processing module, grayscale processing and saving the color face image, performing face segmentation and face Fourier transform on the grayscale face image to form the frequency domain information of the new processed face image in different regions; Step 3: Generate a module that pre-processes the frequency domain information of the face images in different regions in step 1 into blocks to generate different feature image blocks; Step 4: In the calculation module, different feature image blocks enter the GMRF model to extract feature points to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing; Step 5: The merging module combines the GMRF features extracted in step 3, labels the face categories, and packages them together with the feature vectors in step 2; Step 6: Output module, put the feature vectors and labels packaged in step 4 into the SVM classifier for training and recognition, and obtain the classification results by comparing with other multi-angle algorithms; In step 6, the feature vector and SVM classification algorithm are used to classify multi-angle face images, and the classification accuracy is compared by using the influence of GMRF models of different orders on the classification. The steps to determine whether the classification result is correct are as follows: Use SVM multi-classification algorithm to classify multi-angle face images based on feature vectors; Then use the average accuracy results for evaluation. The average accuracy result MAP is:

2. The computer device for face recognition according to claim 1, characterized in that: In the step 2, the step of extracting frequency domain information after graying the color face image is as follows: graying the face image and converting it into a double-precision two-dimensional array, and then performing Fourier transform on the converted two-dimensional array.

3. The computer device for face recognition according to claim 1, characterized in that: In the step 3, the step of dividing the characteristic image into blocks is as follows: reading the frequency domain image features after Fourier transformation, and respectively calculating the length and width values ​​of the characteristic image; The location of the block points of each method is calculated according to the size of different feature image blocks. After the calculation, the processed feature images are taken and saved according to different blocks.

4. The computer device for face recognition according to claim 1, characterized in that: In the step 4, the step of forming the GMRF features after block processing is as follows: first read the different feature image blocks saved in step 3; Then, the feature points of different feature image blocks are extracted using GMRF features to form feature point vectors with calculations.

5. The computer device for face recognition according to claim 4, characterized in that: In the fourth step, after the feature vectors of each feature image block are obtained, they are concatenated together to form a feature vector of the entire feature map and packaged together with the label of the feature map to prepare for subsequent training.

6. A computer device for implementing the face recognition method of claim 1, the device comprising: Collection and import module, used to capture or import color face images; Grayscale processing module, used for grayscale processing module, grayscale processing and saving of color face images, face segmentation and face Fourier transform of the grayscale face images, to form frequency domain information of new processed face images of different regions; A generation module is used to pre-process the frequency domain information of the face image in different regions in step 1 into blocks to generate different feature image blocks; The calculation module is used to extract feature points from different feature image blocks into the GMRF model to form feature vectors to be estimated, and then calculate the variance and mean information to form the GMRF features after block processing; The merging module is used to combine the GMRF features extracted in step 3, label the face categories, and package them together with the feature vector in step 2; The output module is used to input the feature vectors and labels packaged in step 4 into the SVM classifier for training and recognition, and obtain the classification results by comparing with other multi-angle algorithms.

Images