CN115995110A - Face recognition method based on improved PCA+SVC - Google Patents

Abstract

The present disclosure relates to an improved PCA+SVC-based face recognition method, device, electronic equipment and storage medium. Wherein, the method includes: inputting the face image information into the center of the face image to read the image in the form of a two-dimensional grayscale matrix, and converting it into a one-dimensional data sequence; randomly cutting and changing the color of the one-dimensional data sequence , horizontal flipping, vertical flipping, adding noise, affine transformation processing, generating a data-enhanced one-dimensional data sequence; performing data feature extraction on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, Generate facial features; perform parallel identification and classification processing on the facial features based on the SVC method based on Gaussian kernel function, polynomial kernel function, and Sigmoid kernel function respectively, and conduct centralized arbitration based on preset weights to generate the final decision result, and complete face recognition . The present disclosure uses the improved PCA+SVC to perform face recognition, thereby improving the accuracy of face recognition.

Description

A face recognition method based on improved PCA+SVC

Technical Field

The present disclosure relates to the field of face recognition, and in particular, to a face recognition method, device, electronic device and computer-readable storage medium based on improved PCA+SVC.

Background Art

Face recognition is a biometric technology that uses facial information streams collected by cameras to extract features for identity recognition. As a key research project in the field of biometric recognition, face recognition has penetrated into all aspects of people's production and life, from mobile face-swiping login, convenient payment to smart city construction, and its application is becoming more and more extensive. From a technical point of view, under specific experimental conditions, face recognition has achieved an accuracy rate of more than 99%; in actual application scenarios, due to interference from factors such as occlusion and light intensity, the accuracy rate is reduced, but it can also reach an accuracy rate of more than 95%.

At present, the mainstream face recognition technology adopts the face recognition algorithm based on appearance, which extracts global information from the face area as a whole to identify the face. It can be further divided into linear methods and nonlinear methods according to different feature extraction methods. For linear methods, one of the most famous methods is the "Eigenface" proposed by Turk et al., which uses principal component analysis (PCA) to linearly project the image into the low-dimensional space of the training set image, extracts the eigenvector of the face to form the feature space (eigenface). For an image, it is compared with the extracted feature space to identify the face and determine the identity. Yang et al. proposed two-dimensional principal component analysis (2DPCA), which constructs the image covariance matrix based on the two-dimensional image to extract the eigenvector. For nonlinear methods, Kim et al. proposed kernel principal component analysis, which uses a polynomial kernel to map the face image to the feature space through nonlinear mapping, and then uses multiple linear support vector machines (SVM) as face recognizers, and combines the output of each SVM for arbitration to produce the final decision. With the improvement of computer processing power, face recognition technology based on deep learning can automatically learn facial features of the face through the network and recognize the face. However, neural networks are composed of a large number of neurons, which has the disadvantages of requiring a large amount of training data, taking a long time to train, and being difficult to converge.

The existing face recognition technology process mainly consists of the following steps:

(1) Input face image: (a) Read the face image in the form of a two-dimensional grayscale matrix; (b) Convert the grayscale matrix into one dimension to obtain input data.

(2) Use PCA method to extract facial features: (a) Calculate the mean vector of input data; (b) Subtract the mean data from the input data and normalize the data around the mean data; (c) Calculate and decompose the covariance matrix to obtain eigenvectors sorted in descending order of eigenvalues; (d) Take the first k eigenvectors to obtain eigenfaces, calculate the projection of the normalized data in the eigenface space, and extract facial features.

(3). SVC recognition and classification: (a) Select the kernel function to map the extracted facial features into a high-dimensional space; (b) Use grid search to obtain the optimal penalty coefficient and other parameters; (c) Determine the optimal decision boundary to obtain the face image recognition result.

(4). Output recognition results.

The existing technology mainly has the following disadvantages:

(1) Data centering is affected by extreme values. In actual application scenarios, face image collection is easily affected by occlusion, light intensity, and other factors, which can lead to inconsistency, noise, incompleteness (such as missing important attributes) and other problems in the collected data. When using the PCA method to perform consistency reduction on data, it is necessary to calculate the mean vector of the data in the data set and perform data centering. Since the mean vector is easily affected by extreme values, simply using the method of subtracting the mean vector to perform data centering will reduce the interpretability of the extracted feature space and affect the accuracy of face recognition.

(2) SVC relies on kernel function selection. When training SVC to recognize face images, the choice of kernel function affects the performance of the SVC algorithm. Therefore, it is necessary to try multiple kernel functions based on the features of the collected face images to achieve good face recognition results.

(3) Grid search has low computational efficiency. Grid search divides the parameter range to be searched into grids according to the proposed coordinate system to form parameter combinations. The global optimal solution is found by traversing the parameter combinations. This comes at the cost of generating a large number of unnecessary and invalid calculations.

Therefore, one or more methods are needed to solve the above problems.

It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field.

Summary of the invention

The purpose of the present disclosure is to provide a face recognition method, device, electronic device and computer-readable storage medium based on improved PCA+SVC, thereby overcoming one or more problems caused by the limitations and defects of related technologies at least to a certain extent.

According to one aspect of the present disclosure, a face recognition method based on improved PCA+SVC is provided, comprising:

Receiving facial image information, inputting the facial image information into the facial image center to read the image in the form of a two-dimensional grayscale matrix, so as to convert the facial image information into a one-dimensional data sequence, and sending the one-dimensional data sequence to a data enhancement module or a facial feature extraction module;

After receiving the one-dimensional data sequence, the data enhancement module randomly crops, changes color, flips horizontally, flips vertically, adds noise, and performs affine transformation on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

After receiving the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, the facial feature extraction module performs data feature extraction on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, generates facial features, and sends the facial features to the recognition and classification module;

After receiving the facial features, the recognition and classification module performs parallel recognition and classification processing on the facial features based on the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, and generates classification results of the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, and performs centralized arbitration based on preset weights to produce a final decision result to complete face recognition.

In an exemplary embodiment of the present disclosure, the method further includes:

Receive facial image information, input the facial image information into the facial image center to read the image in the form of a two-dimensional grayscale matrix, so as to convert the facial image information into a one-dimensional data sequence, and send the one-dimensional data sequence to a data enhancement module based on a preset probability ρ, and send the one-dimensional data sequence to a facial feature extraction module based on a preset probability 1-ρ;

After receiving the one-dimensional data sequence, the data enhancement module randomly crops, changes color, flips horizontally, flips vertically, adds noise, and performs affine transformation on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

After receiving the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, the facial feature extraction module performs data feature extraction on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, generates facial features, and sends the facial features to the recognition and classification module.

In an exemplary embodiment of the present disclosure, the method further includes:

After receiving the one-dimensional data sequence of size M×N, the data enhancement module randomly crops the one-dimensional data sequence of size M×N and resamples it to data of size M×N;

or randomly performing color change processing on the data by changing the gray value of the data;

or randomly performing horizontal flipping and vertical flipping processing on the data to achieve enlargement of the face image;

Or randomly adding noise of a random value matrix sampled by Gaussian distribution to the data to enhance the classification learning effect;

Or randomly performing a linear transformation of the data from two-dimensional coordinate points to two-dimensional coordinate points to complete an affine transformation process;

A data-enhanced one-dimensional data sequence is generated, and the data-enhanced one-dimensional data sequence is sent to a facial feature extraction module.

In an exemplary embodiment of the present disclosure, the method extracts data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method to generate facial features, including:

Determine the number of grouping segments I based on the size of n one-dimensional data sequences {t ₁ , t ₂ , ... t _n } of the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, and obtain a statistical histogram of the data enhancement, where t _i is a one-dimensional sequence of length M×N;

分配权重w_ij,其中n_ij为该组段像素点个数，n为输入样本的个数；For pixel point (1, j), count the number of pixels in each group of n samples. For group i, according to

Assign weight w _ij , where n _ij is the number of pixels in the segment and n is the number of input samples;

得到加权均值向量

其中X_ij为组段i像素点的像素值之和；according to

Get the weighted mean vector

Where _Xij is the sum of the pixel values of the pixel points in group i;

对输入数据进行中心化处理，得到{t₁，t₂，...t_n}_new，t_j为第j个训练样本；according to

Centralize the input data to obtain {t ₁ , t ₂ , ... t _n } _new , where t _j is the jth training sample;

计算，得到协方差矩阵

For the data {t ₁ , t ₂ , ... t _n } _new processed in the data set, use the formula

Calculate and get the covariance matrix

Using singular value decomposition (SVD), solve all eigenvalues {λ ₁ ,λ ₂ ,...λ _K } and corresponding eigenvectors {V ₁ ,V ₂ ,...V _K } of A ^T A;

构成特征脸空间{e₁，e₂，...e_k}；Sort the feature vectors {V ₁ , V ₂ , ...V _K } in descending order according to the eigenvalue, select the first k feature vectors {V ₁ , V ₂ , ...V _k } according to the face recognition accuracy requirement, and use the formula

Construct the eigenface space {e ₁ , e ₂ , ...e _k };

The data {t ₁ , t ₂ , ...t _n } _new that has been processed in the data set is multiplied by the eigenface space {e ₁ , e ₂ , ...e _k } to obtain the feature representation {t ₁ , t ₂ , ...t _n } _final after dimensionality reduction, and is sent to the recognition and classification module.

In an exemplary embodiment of the present disclosure, the SVC method based on the Gaussian kernel function in the method further includes:

惩罚系数、gamma值参数范围，所述惩罚系数范围为[0,1]；S5.1. Determine the kernel function of SVC

Penalty coefficient, gamma value parameter range, the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S5.2. Based on the Gaussian kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S5.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S5.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_i。S5.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S5.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S5.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _i is output.

In an exemplary embodiment of the present disclosure, the SVC method based on the polynomial kernel function in the method further includes:

S6.1. Determine the parameter ranges of the kernel function k(x, x′)=(xx′+c) ^d , the penalty coefficient, the gamma value, etc. of the SVC, wherein the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S6.2. Based on the polynomial kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S6.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S6.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_j。S6.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S6.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S6.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _j is output.

In an exemplary embodiment of the present disclosure, the SVC method based on the Sigmoid kernel function in the method further includes:

S7.1. Determine the parameter ranges of the kernel function k(x, x′)=tanh(kxx′+c), penalty coefficient, gamma value, etc. of the SVC, wherein the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S7.2. Based on the Sigmoid kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }, and the objective function is optimized.

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S7.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S7.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_k。S7.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S7.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S7.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _k is output.

In an exemplary embodiment of the present disclosure, the method further includes:

After receiving the facial features, the recognition and classification module performs parallel recognition and classification processing on the facial features based on the SVC method of Gaussian kernel function, the SVC method based on polynomial kernel function, and the SVC method based on Sigmoid kernel function, respectively, and generates classification results _Ri of the SVC method based on Gaussian kernel function, _Rj of the SVC method based on polynomial kernel function, and _Rk of the SVC method based on Sigmoid kernel function;

If the classification result R _i of the SVC method based on the Gaussian kernel function, the classification result R _j of the SVC method based on the polynomial kernel function, and the classification result R _k of the SVC method based on the Sigmoid kernel function are the same, any one of the classification results R _i of the SVC method based on the Gaussian kernel function, the classification result R _j of the SVC method based on the polynomial kernel function, and the classification result R _k of the SVC method based on the Sigmoid kernel function is used as the decision result to complete face recognition;

If two of the classification results of the SVC method based on the Gaussian kernel function R _i , the classification results of the SVC method based on the polynomial kernel function R _j , and the classification results of the SVC method based on the Sigmoid kernel function R _k are the same, the two same classification results are aggregated based on the aggregation function as the decision result to complete face recognition;

If the SVC classification result R _i based on the Gaussian kernel function, the SVC classification result R _j based on the polynomial kernel function, and the SVC classification result R _k based on the Sigmoid kernel function are all different, then based on the preset weights ω _i , ω _j , ω _k , ω _i +ω _j +ω _k = 1, centralized arbitration is performed to produce the final decision result, and face recognition is completed.

In one aspect of the present disclosure, a face recognition device based on improved PCA+SVC is provided, comprising:

An input face image center module is used to receive face image information, input the face image information into the face image center to read the image in the form of a two-dimensional grayscale matrix, so as to convert the face image information into a one-dimensional data sequence, and send the one-dimensional data sequence to a data enhancement module;

A data enhancement module is used to randomly crop, change color, flip horizontally, flip vertically, add noise, and perform affine transformation on the one-dimensional data sequence after receiving the one-dimensional data sequence to generate a data-enhanced one-dimensional data sequence, and send the data-enhanced one-dimensional data sequence to the facial feature extraction module;

A facial feature extraction module is used for, after receiving the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, extracting data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, generating facial features, and sending the facial features to the recognition and classification module;

The recognition and classification module is used for, after receiving the facial features, performing parallel recognition and classification processing on the facial features based on the SVC method of Gaussian kernel function, the SVC method based on polynomial kernel function, and the SVC method based on Sigmoid kernel function, respectively, generating classification results of the SVC method based on Gaussian kernel function, the SVC method based on polynomial kernel function, and the SVC method based on Sigmoid kernel function, respectively, and performing centralized arbitration based on preset weights to generate a final decision result, thereby completing face recognition;

The output recognition result center module is used to receive the face recognition result sent by the recognition classification module and print out the face recognition result.

In one aspect of the present disclosure, there is provided an electronic device, comprising:

Processor; and

A memory having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by the processor, implement the method according to any one of the above items.

In one aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method according to any one of the above items is implemented.

In an exemplary embodiment of the present disclosure, a face recognition method based on improved PCA+SVC is disclosed, wherein the method includes: inputting face image information into the center of the face image to read the image in the form of a two-dimensional grayscale matrix and converting it into a one-dimensional data sequence; randomly cropping, color changing, horizontally flipping, vertically flipping, adding noise, and affine transformation processing the one-dimensional data sequence to generate a data-enhanced one-dimensional data sequence; extracting data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method to generate facial features; performing parallel recognition and classification processing on the facial features based on the Gaussian kernel function, the polynomial kernel function, and the Sigmoid kernel function-based SVC method, and performing centralized arbitration based on preset weights to generate a final decision result to complete face recognition. The present disclosure improves the accuracy of face recognition by performing face recognition through improved PCA+SVC.

It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and other features and advantages of the present disclosure will become more apparent by describing in detail example embodiments thereof with reference to the attached drawings.

1A-1B show a flow chart of a face recognition method based on improved PCA+SVC according to an exemplary embodiment of the present disclosure;

2A-2B show a schematic diagram of a facial feature extraction module of a face recognition method based on an improved PCA+SVC according to an exemplary embodiment of the present disclosure;

3A-3E are schematic diagrams showing a recognition and classification module of a face recognition method based on an improved PCA+SVC according to an exemplary embodiment of the present disclosure;

FIG4 shows a schematic block diagram of a face recognition device based on improved PCA+SVC according to an exemplary embodiment of the present disclosure;

FIG5 schematically shows a block diagram of an electronic device according to an exemplary embodiment of the present disclosure; and

FIG. 6 schematically shows a schematic diagram of a computer-readable storage medium according to an exemplary embodiment of the present disclosure.

DETAILED DESCRIPTION

Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, materials, devices, steps, etc. may be adopted. In other cases, known structures, methods, devices, implementations, materials or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

The block diagrams shown in the accompanying drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or these functional entities or parts of functional entities can be implemented in one or more software hardened modules, or these functional entities can be implemented in different networks and/or processor devices and/or microcontroller devices.

In this exemplary embodiment, a face recognition method based on improved PCA+SVC is first provided; referring to FIG. 1A-1B , the face recognition method based on improved PCA+SVC may include the following steps:

Step S110, receiving facial image information, inputting the facial image information into the facial image center to read the image in the form of a two-dimensional grayscale matrix, so as to convert the facial image information into a one-dimensional data sequence, and sending the one-dimensional data sequence to a data enhancement module or a facial feature extraction module;

Step S120, after receiving the one-dimensional data sequence, the data enhancement module randomly crops, changes color, flips horizontally, flips vertically, adds noise, and performs affine transformation on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

Step S130, after receiving the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, the facial feature extraction module performs data feature extraction on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, generates facial features, and sends the facial features to the recognition and classification module;

Step S140, after the recognition and classification module receives the facial features, it performs parallel recognition and classification processing on the facial features based on the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, and generates classification results of the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, and performs centralized arbitration based on preset weights to produce a final decision result, thereby completing face recognition.

In an exemplary embodiment of the present disclosure, a face recognition method based on improved PCA+SVC is disclosed, wherein the method includes: inputting face image information into the center of the face image to read the image in the form of a two-dimensional grayscale matrix and converting it into a one-dimensional data sequence; randomly cropping, color changing, horizontally flipping, vertically flipping, adding noise, and affine transformation processing the one-dimensional data sequence to generate a data-enhanced one-dimensional data sequence; extracting data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method to generate facial features; performing parallel recognition and classification processing on the facial features based on the Gaussian kernel function, the polynomial kernel function, and the Sigmoid kernel function-based SVC method, and performing centralized arbitration based on preset weights to generate a final decision result to complete face recognition. The present disclosure improves the accuracy of face recognition by performing face recognition through improved PCA+SVC.

Next, a face recognition method based on improved PCA+SVC in this exemplary embodiment will be further described.

In step S110, facial image information can be received, and the facial image information can be input into the facial image center to read the image in the form of a two-dimensional grayscale matrix to convert the facial image information into a one-dimensional data sequence, and the one-dimensional data sequence can be sent to a data enhancement module or a facial feature extraction module.

In the embodiment of this example, the method further comprises:

Receive facial image information, input the facial image information into the facial image center to read the image in the form of a two-dimensional grayscale matrix, so as to convert the facial image information into a one-dimensional data sequence, and send the one-dimensional data sequence to a data enhancement module based on a preset probability ρ, and send the one-dimensional data sequence to a facial feature extraction module based on a preset probability 1-ρ;

After receiving the one-dimensional data sequence, the data enhancement module randomly crops, changes color, flips horizontally, flips vertically, adds noise, and performs affine transformation on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

After receiving the data-enhanced one-dimensional data sequence or directly receiving the one-dimensional data sequence, the facial feature extraction module performs data feature extraction on the data-enhanced one-dimensional data sequence or the one-dimensional data sequence based on the PCA method, generates facial features, and sends the facial features to the recognition and classification module.

In the embodiment of this example, the input face image center processes the face image into a one-dimensional grayscale matrix and sends it to the data enhancement module with probability ρ, or directly sends it to the facial feature extraction module with probability 1-ρ, and the probability ρ∈[0,1]. The receiving end of the facial feature extraction module is simultaneously connected to the output end of the data enhancement module and the output end of the input face image center to extract facial features. The facial features extracted by the facial feature extraction module are sent to the recognition and classification module for processing, and the recognition and classification results of the face are obtained through the decision center and sent to the output recognition result center for output. For a face image of size M×N, the input face image center reads the image in the form of a two-dimensional grayscale matrix, flattens it into a one-dimensional sequence of length M×N, and sends it to the data enhancement module with probability ρ.

In step S120, after the data enhancement module receives the one-dimensional data sequence, the one-dimensional data sequence can be randomly cropped, color changed, horizontally flipped, vertically flipped, noise added, and affine transformed to generate a data enhanced one-dimensional data sequence, and the data enhanced one-dimensional data sequence is sent to the facial feature extraction module.

In the embodiment of this example, the method further comprises:

After receiving the one-dimensional data sequence of size M×N, the data enhancement module randomly crops the one-dimensional data sequence of size M×N and resamples it to data of size M×N;

or randomly performing color change processing on the data by changing the gray value of the data;

or randomly performing horizontal flipping and vertical flipping processing on the data to achieve enlargement of the face image;

Or randomly adding noise of a random value matrix sampled by Gaussian distribution to the data to enhance the classification learning effect;

Or randomly performing a linear transformation of the data from two-dimensional coordinate points to two-dimensional coordinate points to complete an affine transformation process;

A data-enhanced one-dimensional data sequence is generated, and the data-enhanced one-dimensional data sequence is sent to a facial feature extraction module.

In the embodiment of this example, the data enhancement module includes 6 data enhancement methods, namely: (1) random cropping (2) color change (3) horizontal flip (4) vertical flip (5) adding noise (6) affine transformation. The image entering the data enhancement module randomly selects one of the methods to change.

(1) Random cropping. The image area is randomly cropped and resampled to M×N size. This method can retain the main features of the image and crop the background noise, thereby reducing the overfitting effect caused by noise.

(2) Color change: Data enhancement is achieved by changing the grayscale value of the image.

(3) Horizontal flip. Rotate the image horizontally to achieve image enlargement.

(4) Vertical flip: Rotate the image vertically to achieve image enlargement.

(5) Add noise. Add a random value matrix sampled from a Gaussian distribution to the face image to help the classifier learn more powerful functions by adding noise to the image.

(6) Affine transformation. A linear transformation from two-dimensional coordinate points to two-dimensional coordinate points can maintain the "parallelism" and "straightness" of the image.

In step S130, after the facial feature extraction module receives the one-dimensional data sequence or the data enhanced one-dimensional data sequence, it can perform data feature extraction on the one-dimensional data sequence or the data enhanced one-dimensional data sequence based on the PCA method, generate facial features, and send the facial features to the recognition and classification module.

In the embodiment of this example, the method extracts data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method to generate facial features, including:

Determine the number of grouping segments I based on the size of n one-dimensional data sequences {t ₁ , t ₂ , ... t _n } of the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, and obtain a statistical histogram of the data enhancement, where t _i is a one-dimensional sequence of length M×N;

分配权重w_ij,其中n_ij为该组段像素点个数，n为输入样本的个数；For pixel point (1, j), count the number of pixels in each group of n samples. For group i, according to

Assign weight w _ij , where n _ij is the number of pixels in the segment and n is the number of input samples;

得到加权均值向量

其中X_ij为组段i像素点的像素值之和；according to

Get the weighted mean vector

Where _Xij is the sum of the pixel values of the pixel points in group i;

对输入数据进行中心化处理，得到{t₁，t₂，...t_n}_new，t_j为第j个训练样本；according to

Centralize the input data to obtain {t ₁ , t ₂ , ... t _n } _new , where t _j is the jth training sample;

计算，得到协方差矩阵

For the data {t ₁ , t ₂ , ... t _n } _new processed in the data set, use the formula

Calculate and get the covariance matrix

Using singular value decomposition (SVD), solve all eigenvalues {λ ₁ ,λ ₂ ,...λ _K } and corresponding eigenvectors {V ₁ ,V ₂ ,...V _K } of A ^T A;

构成特征脸空间{e₁，e₂，...e_k}；Sort the feature vectors {V ₁ , V ₂ , ...V _K } in descending order according to the eigenvalue, select the first k feature vectors {V ₁ , V ₂ , ...V _k } according to the face recognition accuracy requirement, and use the formula

Construct the eigenface space {e ₁ , e ₂ , ...e _k };

The data {t ₁ , t ₂ , ...t _n } _new that has been processed in the data set is multiplied by the eigenface space {e ₁ , e ₂ , ...e _k } to obtain the feature representation {t ₁ , t ₂ , ...t _n } _final after dimensionality reduction, and is sent to the recognition and classification module.

In the embodiment of this example, as shown in FIG2A , the facial feature extraction module mainly uses an improved PCA method to extract features from the data enhancement module or the center of the input face image and perform dimensionality reduction processing. The main steps are:

(1) Statistical histogram. According to the size of the input data {t ₁ , t ₂ , ... t _n }, the number of grouping segments I is determined to obtain the statistical histogram of the input data. _{t i} is a one-dimensional sequence of length M×N.

分配权重w_ij,其中n_ij为该组段像素点个数，n为输入样本的个数。(2) Calculate the weight. For pixel point (1, j), count the number of pixels in each group of n samples. For group i,

Assign weights w _ij , where n _ij is the number of pixels in the segment and n is the number of input samples.

得到加权均值向量

其中X_ij为组段i像素点的像素值之和。(3) Get the mean vector. According to

Get the weighted mean vector

Where _Xij is the sum of the pixel values of the pixels in group i.

对输入数据进行中心化处理，得到{t₁，t₂，...t_n}_new，t_j为第j个训练样本。(4) Centralized processing.

The input data is centralized to obtain {t ₁ , t ₂ , ... t _n } _new , where t _j is the jth training sample.

计算，得到协方差矩阵

(5) Calculate the covariance matrix. The covariance matrix represents the relationship between any two pixels in the image, so for the data {t ₁ , t ₂ , ... t _n } _new after data processing, according to the formula

Calculate and get the covariance matrix

(6) Decomposition of the covariance matrix. Since the dimension of the face image is large, singular value decomposition (SVD) is used to solve all the eigenvalues {λ ₁ ,λ ₂ ,...λ _K } of ^AT A and the corresponding eigenvectors {V ₁ ,V ₂ ,...V _K }.

构成特征脸空间{e₁，e₂，...e_k]。图2B为空间中几例特征向量的可视化结果。(7) Sort the feature vectors {V ₁ , V ₂ , ... V _K } in descending order of eigenvalues, select the first k feature vectors {V ₁ , V ₂ , ... V _k } according to the face recognition accuracy requirements, and use the formula

The eigenface space {e ₁ , e ₂ , ... e _k ] is constructed. FIG2B is the visualization result of several eigenvectors in the space.

(8) Multiply the data {t ₁ , t ₂ , ... t _n } _new after data center processing with the eigenface space {e ₁ , e ₂ , ... e _k } to obtain the feature representation {t ₁ , t ₂ , ... t _n } _final after dimensionality reduction, and send it to the recognition and classification module.

In step S140, after the recognition and classification module receives the facial features, it can perform parallel recognition and classification processing on the facial features based on the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, to generate classification results of the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function, respectively, and perform centralized arbitration based on preset weights to produce a final decision result, thereby completing face recognition.

In the embodiment of this example, as shown in FIG3A , the recognition and classification module mainly uses an improved integrated SVC method to process the feature representation output {t ₁ , t ₂ , ... t _n } _final of the facial feature extraction module, and obtains the classification and recognition result through the decision center. The module mainly integrates three types of SVC, namely (1) SVC based on Gaussian kernel function (2) SVC based on polynomial kernel function and (3) SVC based on Sigmoid kernel function. For the same feature representation, the classification results R _i , R _j , R _k are obtained respectively, and the three classification results are multiplied by weights ω _i , ω _j , ω _k respectively, and the final decision result is generated by centralized arbitration. The weights ω _i , ω _j , ω _k are obtained by dividing the number of correct classifications of the corresponding SVC C _s by the total number of correct classifications C, ω _i + ω _j + ω _k = 1.

In the embodiment of this example, the SVC method based on the Gaussian kernel function in the method further includes:

惩罚系数、gamma值参数范围，所述惩罚系数范围为[0,1]；S5.1. Determine the kernel function of SVC

Penalty coefficient, gamma value parameter range, the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S5.2. Based on the Gaussian kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S5.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S5.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_i。S5.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S5.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S5.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _i is output.

作为其核函数，可以对任意的数据映射成线性可分，对于没有先验知识的特征表示，基于高斯核的SVC具有良好的性能，核值范围为[0,1]。如图3B所示，其计算过程为：In the embodiment of this example, the Gaussian kernel-based SVC uses

As its kernel function, any data can be mapped into linear separable. For feature representation without prior knowledge, SVC based on Gaussian kernel has good performance, and the kernel value range is [0,1]. As shown in Figure 3B, its calculation process is:

(a) Determine the range of parameters such as the penalty coefficient and gamma value of SVC. The penalty coefficient can balance the misclassified samples and the classification interval and control the penalty intensity. The range is [0,1]. The larger the penalty coefficient, the less tolerance the classifier has for noise points within the boundary.

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度。(b) Based on the Gaussian kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₂ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

The decision boundary and classification results are obtained, and the classifier accuracy is calculated based on the classification results using a ten-fold cross validation method.

个参数。对每个组采样k次重复步骤(b)，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤(b)计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_i。(c) Using the improved grid search method, the possible parameter values are arranged and combined according to the specified parameter range to obtain M parameter results. The M parameter results are divided into m groups, each group

Parameters. Sample each group k times and repeat step (b), calculate the average accuracy and sort in descending order to take the top t groups; sample t groups 2k times and repeat step (b) to calculate the average accuracy and sort in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _i is output.

In the embodiment of this example, it is characterized in that the SVC method based on the polynomial kernel function in the method further includes:

S6.1. Determine the parameter ranges of the kernel function k(x, x′)=(xx′+c) ^d , the penalty coefficient, the gamma value, etc. of the SVC, wherein the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S6.2. Based on the polynomial kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S6.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S6.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_j。S6.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S6.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S6.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _j is output.

In the embodiment of this example, the polynomial kernel function based SVC uses k(x, x′)=(xx′+c) ^d as its kernel function, which can solve nonlinear problems by increasing the dimension and allows subjective setting of the power. As shown in FIG3C , the calculation process is:

(a) Determine the range of parameters such as the penalty coefficient and gamma value of SVC. The penalty coefficient can balance the misclassified samples and the classification interval and control the penalty intensity. The range is [0,1]. The larger the penalty coefficient, the less tolerance the classifier has for noise points within the boundary.

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度。(b) Based on the polynomial kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₁ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

The decision boundary and classification results are obtained, and the classifier accuracy is calculated based on the classification results using a ten-fold cross validation method.

个参数。对每个组采样k次重复步骤(b)，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤(b)计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_j。(c) Using the improved grid search method, the possible parameter values are arranged and combined according to the specified parameter range to obtain M parameter results. The M parameter results are divided into m groups, each group

Parameters. Sample each group k times and repeat step (b), calculate the average accuracy and sort in descending order to take the top t groups; sample t groups 2k times and repeat step (b) to calculate the average accuracy and sort in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _j is output.

In the embodiment of this example, the SVC method based on the Sigmoid kernel function in the method further includes:

S7.1. Determine the parameter ranges of the kernel function k(x, x′)=tanh(kxx′+c), penalty coefficient, gamma value, etc. of the SVC, wherein the penalty coefficient range is [0,1];

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度；S7.2. Based on the Sigmoid kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₂ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }, and the objective function is optimized.

Obtain the decision boundary and classification results, and calculate the classifier accuracy based on the classification results using a ten-fold cross validation method;

个参数，对每个组采样k次重复步骤S7.2，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤S7.2计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_k。S7.3. Use the improved grid search method to arrange and combine the parameter values according to the specified parameter range to obtain M parameter results. Divide the M parameter results into m groups, each group

parameters, sample each group k times and repeat step S7.2, calculate the average accuracy and sort them in descending order to take the top t groups; sample t groups 2k times and repeat step S7.2 to calculate the average accuracy and sort them in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _k is output.

In the embodiment of this example, the SVC based on the Sigmoid kernel function uses k(x, x′)=tanh(kxx′+c) as its kernel function, which has good generalization ability for unknown samples and can reduce the occurrence of over-learning. As shown in Figure 3D, its calculation process is:

(a) Determine the range of parameters such as the penalty coefficient and gamma value of SVC. The penalty coefficient can balance the misclassified samples and the classification interval and control the penalty intensity. The range is [0,1]. The larger the penalty coefficient, the less tolerance the classifier has for noise points within the boundary.

得到决策边界及分类结果，根据分类结果以十折交叉验证方式计算分类器精确度。(b) Based on the Sigmoid kernel function with determined parameters, the input feature representation {t ₁ , t ₂ , ... t _n } _final and the corresponding class labels {t′ ₁ , t′ ₂ , ... t′ _n } are spatially mapped to obtain high-dimensional data {k ₁ , k ₂ , ... k _n } and the corresponding class labels {k′ ₁ , k′ ₁ , ... k′ _n }. By optimizing the objective function

The decision boundary and classification results are obtained, and the classifier accuracy is calculated based on the classification results using a ten-fold cross validation method.

个参数。对每个组采样k次重复步骤(b)，计算平均精确度并降序排列取前t个组；对t个组采样2k次重复步骤(b)计算平均精确度并降序排列取前

个组,直到得到最优参数，并输出最终识别分类结果R_k。(c) Using the improved grid search method, the possible parameter values are arranged and combined according to the specified parameter range to obtain M parameter results. The M parameter results are divided into m groups, each group

Parameters. Sample each group k times and repeat step (b), calculate the average accuracy and sort in descending order to take the top t groups; sample t groups 2k times and repeat step (b) to calculate the average accuracy and sort in descending order to take the top

groups until the optimal parameters are obtained, and the final recognition and classification result R _k is output.

In the embodiment of this example, the method further comprises:

After receiving the facial features, the recognition and classification module performs parallel recognition and classification processing on the facial features based on the SVC method of Gaussian kernel function, the SVC method based on polynomial kernel function, and the SVC method based on Sigmoid kernel function, respectively, and generates classification results _Ri of the SVC method based on Gaussian kernel function, _Rj of the SVC method based on polynomial kernel function, and _Rk of the SVC method based on Sigmoid kernel function;

If the classification result R _i of the SVC method based on the Gaussian kernel function, the classification result R _j of the SVC method based on the polynomial kernel function, and the classification result R _k of the SVC method based on the Sigmoid kernel function are the same, any one of the classification results R _i of the SVC method based on the Gaussian kernel function, the classification result R _j of the SVC method based on the polynomial kernel function, and the classification result R _k of the SVC method based on the Sigmoid kernel function is used as the decision result to complete face recognition;

If two of the classification results of the SVC method based on the Gaussian kernel function R _i , the classification results of the SVC method based on the polynomial kernel function R _j , and the classification results of the SVC method based on the Sigmoid kernel function R _k are the same, the two same classification results are aggregated as the decision result based on the aggregation function to complete the face recognition;

If the SVC classification result R _i based on the Gaussian kernel function, the SVC classification result R _j based on the polynomial kernel function, and the SVC classification result R _k based on the Sigmoid kernel function are all different, then based on the preset weights ω _i , ω _j , ω _k , ω _i +ω _j +ω _k = 1, centralized arbitration is performed to produce the final decision result, and face recognition is completed.

In the embodiment of this example, as shown in Figure 3D, the input of the decision center is the output result R _i based on the Gaussian kernel function SVC and the corresponding weight ω _i , the output result R _j based on the polynomial kernel function SVC and the corresponding weight ω _j , and the output result R _k based on the Sigmoid kernel function SVC and the corresponding weight ω _k , and the decision is made according to the following judgment rules.

If _Ri = _Rj = _Rk , that is, the output results of the three types of SVCs are consistent, then _Ri is directly output as the decision result; if two of the output results are consistent, then the aggregation function G is first used to aggregate the same results, complete the voting, and output the decision result corresponding to the largest number of votes; if the output results of the three types of SVCs are different, then according to the weights, the output corresponding to the maximum weight is taken as the decision result.

In the embodiment of this example, the output result center prints out the final decision result R _i or R _j or R _k of the integrated SVC.

In the embodiment of this example, the present disclosure solves the problem that data centering is affected by extreme values. The method of simply summing the data to obtain the mean is easily affected by extreme values, so that the obtained mean is not suitable for describing the concentration of the overall data, resulting in reduced explanatory power of the main component features extracted after the data is centralized. In this regard, we improve the way to obtain the average value, calculate the histogram of the input data, and obtain the weighted mean according to the weight, which greatly reduces the influence of extreme values on the mean, making the subsequent data centering more in line with the actual data and obtaining more accurate feature faces. The present disclosure solves the problem that SVC depends on the selection of kernel functions. When training SVC to recognize face images, the choice of kernel function affects the effect of SVC algorithm, so it is necessary to try multiple kernel functions according to the features of the collected face images, and this process consumes time and energy. In this regard, we propose an integrated SVC, combining the recognition and classification results of Gaussian kernel function, polynomial kernel function, and Sigmoid kernel function, centralized arbitration to produce the final decision, realize the function of the classifier automatically selecting the appropriate kernel function, reduce the time and energy consumed by kernel function selection, and improve the accuracy of face recognition. The present disclosure improves the grid search algorithm. In the process of determining the optimal parameter combination through grid search, a width-first grid search strategy is adopted to continuously prune branches to narrow the search range and reduce the complexity of parameter search. The present invention adds a data enhancement module to generate more diverse training data close to the real distribution using limited training sample data, so as to prompt the model to learn more robust features, improve the generalization ability of the model, and further improve the accuracy of face recognition.

It should be noted that, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and/or one step may be decomposed into multiple steps, etc.

In addition, in this exemplary embodiment, a face recognition device based on improved PCA+SVC is also provided. Referring to FIG4 , the face recognition device 400 based on improved PCA+SVC may include: an input face image center module 410, a data enhancement module 420, a facial feature extraction module 430, a recognition classification module 440, and an output recognition result center module 450.

in:

The face image input center module 410 is used to receive face image information, input the face image information into the face image center to read the image in the form of a two-dimensional gray matrix, so as to convert the face image information into a one-dimensional data sequence, and send the one-dimensional data sequence to the data enhancement module;

The data enhancement module 420 is used to randomly crop, change the color, flip horizontally, flip vertically, add noise, and perform affine transformation on the one-dimensional data sequence after receiving the one-dimensional data sequence to generate a data-enhanced one-dimensional data sequence, and send the data-enhanced one-dimensional data sequence to the facial feature extraction module;

A facial feature extraction module 430 is used for, after receiving the one-dimensional data sequence or the data-enhanced one-dimensional data sequence, extracting data features from the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on the PCA method, generating facial features, and sending the facial features to the recognition and classification module;

The recognition and classification module 440 is used to perform parallel recognition and classification processing on the facial features based on the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function after receiving the facial features, and generate classification results of the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function, and the SVC method based on the Sigmoid kernel function respectively, and perform centralized arbitration based on preset weights to generate a final decision result to complete face recognition;

The output recognition result central module 450 is used to receive the face recognition result sent by the recognition classification module and print out the face recognition result.

The specific details of each of the above-mentioned face recognition device modules based on improved PCA+SVC have been described in detail in the corresponding face recognition method based on improved PCA+SVC, so they will not be repeated here.

It should be noted that, although several modules or units of a face recognition device 400 based on improved PCA+SVC are mentioned in the above detailed description, such division is not mandatory. In fact, according to an embodiment of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as systems, methods or program products. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: complete hardware embodiments, complete software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, which may be collectively referred to herein as "circuits", "modules" or "systems".

The electronic device 500 according to this embodiment of the present invention is described below with reference to Fig. 5. The electronic device 500 shown in Fig. 5 is only an example and should not bring any limitation to the functions and application scope of the embodiment of the present invention.

As shown in Fig. 5, the electronic device 500 is in the form of a general computing device. The components of the electronic device 500 may include but are not limited to: the at least one processing unit 510, the at least one storage unit 520, a bus 530 connecting different system components (including the storage unit 520 and the processing unit 510), and a display unit 540.

The storage unit stores a program code, which can be executed by the processing unit 510, so that the processing unit 510 performs the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of this specification. For example, the processing unit 510 can perform steps S110 to S140 as shown in Figure 1.

The storage unit 520 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 5201 and/or a cache storage unit 5202 , and may further include a read-only storage unit (ROM) 5203 .

The storage unit 520 may also include a program/utility 5204 having a set (at least one) of program modules 5203, such program modules 5205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

Bus 550 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 500 may also communicate with one or more external devices 570 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 500, and/or communicate with any device that enables the electronic device 500 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input/output (I/O) interface 550. Furthermore, the electronic device 500 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and/or public networks, such as the Internet) via a network adapter 560. As shown, the network adapter 560 communicates with other modules of the electronic device 500 via a bus 550. It should be understood that, although not shown in the figure, other hardware and/or software modules may be used in conjunction with the electronic device 500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible embodiments, various aspects of the present invention can also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to perform the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

Referring to FIG6 , a program product 600 for implementing the above method according to an embodiment of the present invention is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (11)

1. A face recognition method based on improved pca+svc, the method comprising:

Receiving face image information, inputting the face image information into a face image center, reading an image in a two-dimensional gray matrix form, converting the face image information into a one-dimensional data sequence, and sending the one-dimensional data sequence to a data enhancement module or a facial feature extraction module;

after receiving a one-dimensional data sequence, the data enhancement module randomly performs random cutting, color change, horizontal overturning, vertical overturning, noise adding and affine transformation processing on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

after receiving the one-dimensional data sequence or the data enhanced one-dimensional data sequence, the facial feature extraction module performs data feature extraction on the one-dimensional data sequence or the data enhanced one-dimensional data sequence based on a PCA method to generate facial features, and sends the facial features to the recognition classification module;

after the recognition classification module receives the facial features, the SVC method based on the Gaussian kernel function, the SVC method based on the polynomial kernel function and the SVC method based on the Sigmoid kernel function respectively carry out parallel recognition classification processing on the facial features, SVC method classification results based on the Gaussian kernel function, SVC method classification results based on the polynomial kernel function and SVC method classification results based on the Sigmoid kernel function are respectively generated, centralized arbitration is carried out based on preset weights to generate final decision results, and face recognition is completed.

2. The method of claim 1, wherein the method further comprises:

receiving face image information, inputting the face image information into a face image center, reading an image in a two-dimensional gray matrix form, converting the face image information into a one-dimensional data sequence, transmitting the one-dimensional data sequence to a data enhancement module based on a preset probability rho, and transmitting the one-dimensional data sequence to a facial feature extraction module based on a preset probability 1-rho;

after receiving a one-dimensional data sequence, the data enhancement module randomly performs random cutting, color change, horizontal overturning, vertical overturning, noise adding and affine transformation processing on the one-dimensional data sequence to generate a data enhanced one-dimensional data sequence, and sends the data enhanced one-dimensional data sequence to the facial feature extraction module;

after the facial feature extraction module receives the one-dimensional data sequence or the data enhanced one-dimensional data sequence, carrying out data feature extraction on the one-dimensional data sequence or the data enhanced one-dimensional data sequence based on a PCA method, generating facial features, and sending the facial features to the recognition classification module.

3. The method of claim 1, wherein the method further comprises:

after receiving a one-dimensional data sequence with the size of MxN, the data enhancement module randomly and randomly cuts the one-dimensional data sequence with the size of MxN, and resamples the one-dimensional data sequence to data with the size of MxN;

or performing a color change process on the data by changing a gray value of the data at random;

or carrying out horizontal overturning and vertical overturning treatment on the data at random to realize the amplification of the face image;

or randomly adding noise of a random value matrix subjected to Gaussian distribution sampling into the data so as to enhance the classification learning effect;

or carrying out linear transformation from the two-dimensional coordinate point to the two-dimensional coordinate point on the data randomly to finish affine transformation processing;

and generating a data enhanced one-dimensional data sequence, and sending the data enhanced one-dimensional data sequence to a facial feature extraction module.

4. The method of claim 1, wherein the method performs data feature extraction on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence based on a PCA method, the generating facial features comprising:

N one-dimensional data sequences { t } based on the one-dimensional data sequence or the data-enhanced one-dimensional data sequence ₁ ，t ₂ ，...t _n Determining the number I of grouping segments to obtain a statistical histogram of the data enhancement, t _i A one-dimensional sequence of length m×n;

for pixel points (1, j), counting the number of pixel points of each group of segments of n samples, and for group segment i, according to the following steps

Assigning weights w _ij Wherein n is _ij N is the number of input samples for the number of the pixel points of the group of segments;

according to

Obtaining a weighted mean vector->

Wherein X is _ij The sum of pixel values of the pixel points of the group section i is set;

according to

The input data is centered to obtain { t } ₁ ，t ₂ ，...t _n } _new ，t _j For the j-th training sample;

to the data after being processed by data centeringData { t } ₁ ，t ₂ ，...t _n } _new According to the formula

Calculating to obtain covariance matrix->

Solving for A by Singular Value Decomposition (SVD) ^T All eigenvalues { lambda } ₁ ，λ ₂ ，...λ _K ' and corresponding feature vector { V } ₁ ，V ₂ ，...V _K }；

The feature vectors { V } are ordered in order of the feature values from large to small ₁ ，V ₂ ，...V _K Selecting the first k eigenvectors { V } according to the face recognition accuracy requirement ₁ ，V ₂ ，...V _k According to the formula }

Constitute the feature face space { e ₁ ，e ₂ ，...e _k }；

Data { t after data centering processing ₁ ，t ₂ ，...t _n } _new And feature face space { e } ₁ ，e ₂ ，...e _k Multiplication to obtain feature representation { t } after dimension reduction ₁ ，t ₂ ，...t _n } _final And sending the identification and classification module.

5. The method of claim 1, wherein the SVC method based on gaussian kernel function in the method further comprises:

s5.1. determining a kernel function of SVC

Punishment coefficient and gamma value parameter range, wherein the punishment coefficient range is [0,1 ]]；

S5.2, representing the input characteristics based on the Gaussian kernel function of the determined parameters{t ₁ ，t ₂ ，...t _n } _final Corresponding class label { t' ₁ ，t′ ₁ ，...t′ _n Space mapping to obtain high-dimensional data { k }, and ₁ ，k ₂ ，...k _n 'and corresponding class label { k' ₁ ，k′ ₂ ，...k′ _n By optimizing an objective function

Obtaining decision boundaries and classification results, and calculating the accuracy of the classifier in a ten-fold cross-validation mode according to the classification results;

s5.3, adopting an improved grid search method, arranging and combining parameter values according to a specified parameter range to obtain M parameter results, and dividing the M parameter results into M groups, wherein each group comprises a plurality of groups of parameters

Sampling k times for each group, repeating the step S5.2, calculating average accuracy, and arranging the first t groups in descending order; the average accuracy is calculated by repeating step S5.2 2 for 2k samples of t groups and the pre +.>

Grouping until obtaining optimal parameters, and outputting final identification classification result R _i 。

6. The method of claim 5, wherein the SVC method based on polynomial kernel function in the method further comprises:

S6.1. determining a kernel function k (x, x ') = (xx' +c) of SVC ^d Parameter ranges such as penalty coefficient, gamma value and the like, wherein the penalty coefficient range is [0,1]；

S6.2, representing the input characteristics { t }, based on the polynomial kernel function of the determined parameters ₁ ，t ₂ ，...t _n } _final Corresponding class label { t' ₁ ，t′ ₁ ，...t′ _n Space mapping to obtain high-dimensional data { k }, and ₁ ，k ₂ ，...k _n 'and corresponding class label { k' ₁ ，k′ ₁ ，...k′ _n By optimizing an objective function

Obtaining decision boundaries and classification results, and calculating the accuracy of the classifier in a ten-fold cross-validation mode according to the classification results;

s6.3, adopting an improved grid search method, arranging and combining parameter values according to a specified parameter range to obtain M parameter results, and dividing the M parameter results into M groups, wherein each group comprises a plurality of groups of parameters

Sampling k times for each group, repeating the step S6.2, calculating average accuracy, and arranging the first t groups in descending order; the average accuracy is calculated by repeating step S6.2 2 for 2k samples of t groups and the pre +.>

Grouping until obtaining optimal parameters, and outputting final identification classification result R _j 。

7. The method of claim 6, wherein the SVC method based on Sigmoid kernel function in the method further comprises:

s7.1, determining parameter ranges such as a kernel function k (x, x ')=tanh (kxx' +c), a penalty coefficient, a gamma value and the like of the SVC, wherein the penalty coefficient range is [0,1];

S7.2, based on the Sigmoid kernel function of the determined parameters, representing the input characteristics { t } ₁ ，t ₂ ，...t _n } _final Corresponding class label { t' ₁ ，t′ ₁ ，...t′ _n Space mapping to obtain high-dimensional data { k }, and ₁ ，k ₂ ，...k _n 'and corresponding class label { k' ₁ ，k′ ₁ ，...k′ _n By optimizing an objective function

Obtaining decision boundaries and classification results, and calculating the accuracy of the classifier in a ten-fold cross-validation mode according to the classification results;

s7.3, adopting an improved grid search method, arranging and combining parameter values according to a specified parameter range to obtain M parameter results, and dividing the M parameter results into M groups, wherein each group comprises a plurality of groups of parameters

Sampling k times for each group, repeating the step S7.2, calculating average accuracy, and arranging the first t groups in descending order; the average accuracy is calculated by repeating step S7.2 2k times for t groups and the pre +.>

Grouping until obtaining optimal parameters, and outputting final identification classification result R _k 。

8. The method of claim 7, wherein the method further comprises:

after the recognition classification module receives the facial features, the recognition classification module performs parallel recognition classification processing on the facial features by using a SVC method based on a Gaussian kernel function, a SVC method based on a polynomial kernel function and a SVC method based on a Sigmoid kernel function, and generates SVC method classification results R based on the Gaussian kernel function respectively _i SVC method classification result R based on polynomial kernel function _j SVC method classification result R based on Sigmoid kernel function _k ；

If the SVC method based on Gaussian kernel function classifies the result R _i SVC method classification result R based on polynomial kernel function _j SVC method classification result R based on Sigmoid kernel function _k The SVC method based on Gaussian kernel function is used for classifying the result R _i SVC method classification result R based on polynomial kernel function _j SVC method classification result R based on Sigmoid kernel function _k Any classification result in the set is used as a decision result to finish face recognition;

if saidSVC method classification result R based on Gaussian kernel function _i SVC method classification result R based on polynomial kernel function _j SVC method classification result R based on Sigmoid kernel function _k If the two classification results are the same, aggregating the same two classification results based on the aggregation function as a decision result to finish face recognition;

if the SVC method based on Gaussian kernel function classifies the result R _i SVC method classification result R based on polynomial kernel function _j SVC method classification result R based on Sigmoid kernel function _k All are different, based on preset weight omega _i ,ω _j ,ω _k Said omega _i +ω _j +ω _k =1, performing centralized arbitration to generate a final decision result, and completing face recognition.

9. A face recognition device based on improved pca+svc, the device comprising:

the input face image center module is used for receiving face image information, inputting the face image information into a face image center, reading an image in a two-dimensional gray matrix form, converting the face image information into a one-dimensional data sequence, and sending the one-dimensional data sequence to the data enhancement module or the facial feature extraction module;

the data enhancement module is used for randomly carrying out random cutting, color change, horizontal overturning, vertical overturning, noise adding and affine transformation processing on the one-dimensional data sequence after receiving the one-dimensional data sequence, generating a data enhanced one-dimensional data sequence, and sending the data enhanced one-dimensional data sequence to the facial feature extraction module;

the facial feature extraction module is used for carrying out data feature extraction on the one-dimensional data sequence or the data enhanced one-dimensional data sequence based on a PCA method after receiving the one-dimensional data sequence or the data enhanced one-dimensional data sequence, generating facial features and sending the facial features to the recognition classification module;

the recognition classification module is used for carrying out parallel recognition classification processing on the facial features respectively based on a SVC method of a Gaussian kernel function, a SVC method of a polynomial kernel function and a SVC method of a Sigmoid kernel function after receiving the facial features, respectively generating a SVC method classification result of the Gaussian kernel function, a SVC method classification result of the polynomial kernel function and a SVC method classification result of the Sigmoid kernel function, carrying out centralized arbitration based on preset weights to generate a final decision result, and completing face recognition;

And the output recognition result center module is used for receiving the face recognition result sent by the recognition classification module and printing and outputting the face recognition result.

10. An electronic device, comprising

A processor; and

a memory having stored thereon computer readable instructions which, when executed by the processor, implement the method according to any of claims 1 to 8.

11. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method according to any of claims 1 to 8.

Images