Face recognition method and system based on image exposure correction network

Abstract

The application discloses a face recognition method and system based on an image exposure correction network. First, the input image is preprocessed, then the preprocessed image is input into a face image extraction network to obtain a face image, then the face image is input into an image exposure correction network to obtain a normal exposure face image, and finally the normal exposure face image is input into a face recognition network to obtain a face recognition result. The image exposure correction network comprises a color space conversion module, an exposure feature extraction network, a brightness mapping network and a detail enhancement module. The exposure feature extraction network comprises a down-sampling layer, a random cropping module and an image encoder A module. The face recognition network comprises an image encoder B based on SimCLR. The application has the advantages of stability, independence, rapidity and efficiency, and can greatly improve the accuracy of face recognition under abnormal lighting conditions.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of face recognition, and relates to a face recognition method and system, in particular to a face recognition method and system based on an image exposure correction network. BACKGROUND

[0002] Since the 1990s, face recognition technology has made great progress. However, during the acquisition and transmission of face images, they may be affected by factors such as changes in lighting, expressions, and poses, which pose great challenges to face recognition.

[0003] In early research, face recognition mainly relied on geometric feature extraction methods, such as describing face features through the distances and angles between facial key points [1]. These methods can to some extent cope with changes in facial expressions and poses, but are limited by computational complexity and recognition accuracy. Subsequently, algebraic methods based on Eigenfaces [2] gradually became popular, which effectively reduced the data dimension and improved the recognition efficiency by extracting the main components of face images through singular value decomposition (SVD). However, these methods still have limitations in dealing with changes in lighting and occlusions. With the development of machine learning, machine learning algorithms have been widely applied in face recognition, and statistical learning-based methods such as support vector machines [3] and boosting algorithms [4] have gradually become mainstream, greatly improving the accuracy and robustness of face recognition.

[0004] In recent years, with the breakthrough of deep learning technology, especially the successful application of convolutional neural networks (CNN) [5] in image recognition, face recognition technology has made revolutionary progress. Deep learning models can automatically learn deep features of face images, effectively overcoming the limitations of traditional methods in dealing with complex scenarios. Deep learning models have shown excellent ability in handling complex scenarios and diverse face images. For example, DeepFace [6] extracts features through deep convolutional networks, achieving efficient recognition on large databases. However, current deep learning-based methods still have limitations in mathematical modeling for face recognition.

[0005] [1] Goldstein, L., Samet, H., & Teller, S. (1989). Geometric and photometric invariance for object recognition. International Journal of Computer Vision, 3(3), 209-226.

[0006] [2] Turk, M., & Pentland, A. (1991). Eigenfaces for recognition. Journal of cognitive neuroscience, 3(1), 71-86.

[0007] [3] Platt, J. C. (1999). Using sparsity to improve support vector machines. In Advances in neural information processing systems (pp. 547-553).

[0008] [4] Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of computer and system sciences, 55(1), 119-139.

[0009] [5] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).

[0010] [6] Taigman, Y., Yang, M., Ranzato, M., & Wolf, L. (2014). DeepFace: Closing the gap to human-level performance in face verification. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1701-1708). SUMMARY

[0011] In order to solve the problem of low accuracy of the current face recognition model under the condition of abnormal light condition, the application provides a face recognition method and system based on an image exposure correction network, which enhances the accuracy of the face recognition model through the design of the image exposure correction network.

[0012] According to an aspect of the present application, a face recognition method based on an image exposure correction network is provided, comprising:

[0013] Input the image to be detected into a face image extraction network to obtain a face image; the face image extraction network comprises a HOG feature extraction module;

[0014] Input the obtained face image into an image exposure correction network to obtain a normal exposure face image; the image exposure correction network comprises a color space conversion module, an exposure feature extraction network, a brightness mapping network and a detail enhancement module;

[0015] Input the obtained normal exposure face image into a face recognition network to obtain a face recognition result; the face recognition network comprises a SimCLR-based image encoder.

[0016] As a further technical solution, the method further comprises:

[0017] Obtain an image to be detected and pre-process the image to be detected.

[0018] As a further technical solution, the exposure feature extraction network comprises a down-sampling layer, a random cropping module and an image encoder module; the down-sampling layer comprises 4 basic network structures and 1 sampling layer, each basic network structure comprises 2 residual blocks and 1 self-attention layer; the random cropping module is used to provide images with fixed resolution; the image encoder module comprises a residual neural network and 1 self-attention layer.

[0019] As a further technical solution, the training of the exposure feature extraction network comprises:

[0020] Import an exposure error image dataset;

[0021] Input the exposure error image and the normal exposure image into the same down-sampling layer respectively;

[0022] Input the down-sampled exposure error image and normal exposure image and the original exposure error image and normal exposure image into the random cropping module respectively;

[0023] Input the randomly cropped image into the image encoder module, and then project the output of the image encoder module to the feature space through the decoder;

[0024] The training data is input into the image encoder module and the decoder network for training, and the model parameters are constantly optimized through a standardized temperature scaled cross-entropy loss function, and an exposure feature extraction network trained is output.

[0025] As a further technical solution, the training of the brightness mapping network comprises:

[0026] Importing an exposure error image dataset;

[0027] The exposure error image and the exposure normal image are trained on the trained frozen exposure feature extraction network using a regularized linear regressor to obtain a brightness mapping relationship.

[0028] As a further technical solution, the implementation of the detail enhancement module comprises:

[0029] A Gaussian filter is used to perform smoothing filtering on the brightness mapped image to obtain a detail enhanced image.

[0030] As a further technical solution, the training of the face recognition network comprises:

[0031] Importing a face recognition dataset;

[0032] Random image enhancement is performed on the face image;

[0033] The enhanced face image is input into an image encoder trained based on SimCLR, and the model parameters are constantly optimized through gradient descent and back propagation to obtain a contrast pre-training module, wherein the image encoder comprises one basic network structure, the basic network structure comprises two residual blocks and one self-attention layer, and each residual block is composed of two convolutional layers.

[0034] According to an aspect of the present application, a face recognition system based on an image exposure correction network is provided, comprising:

[0035] A face image extraction unit is configured to input a to-be-detected image into a face image extraction network to obtain a face image, and the face image extraction network comprises a HOG feature extraction module;

[0036] An image exposure correction unit is configured to input the obtained face image into an image exposure correction network to obtain an exposure normal face image, and the image exposure correction network comprises a color space conversion module, an exposure feature extraction network, a brightness mapping network and a detail enhancement module;

[0037] A face recognition unit is configured to input the obtained exposure normal face image into a face recognition network to obtain a face recognition result, and the face recognition network comprises an image encoder based on SimCLR.

[0038] According to an aspect of the present specification, there is provided a face recognition device based on an image exposure correction network, comprising: one or more processors; and a storage device for storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the steps of the face recognition method based on the image exposure correction network.

[0039] According to an aspect of the present specification, there is provided a non-transitory computer-readable storage medium storing computer instructions that cause the computer to perform the steps of the face recognition method based on the image exposure correction network.

[0040] Compared with the prior art, the present application has the following advantages:

[0041] The present application uses a face recognition model based on an image correction network to perform face recognition on images with abnormal exposure. This technology combines image inpainting and deep learning algorithms. First, the HOG is used to obtain the face image features contained in the original input image. Then, the face image is converted to HSV color space from RGB color space. Then, the exposure feature extraction network is used to extract the features of the face image. Then, the brightness mapping network is used to map the extracted face image feature data to obtain an exposure normal image with missing details. Then, the detail enhancement network model is used to enhance the details of the exposure normal image with missing details to obtain a detail-enhanced exposure normal face image. Finally, the face recognition network is used to perform face recognition on the detail-enhanced exposure normal face image to obtain the face recognition result. Compared with existing face recognition models, the present application has the advantages of stability, independence, rapidity and efficiency, and can greatly improve the accuracy of face recognition under abnormal lighting conditions. BRIEF DESCRIPTION OF DRAWINGS

[0042] To make the technical solutions in the embodiments of the present application or the prior art clearer, the drawings used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 The method flowchart of the embodiments of the present application;

[0044] Figure 2 The image exposure correction network structure diagram of the embodiments of the present application;

[0045] Figure 3 The exposure correction network training flowchart of the embodiments of the present application;

[0046] Figure 4 A luminance mapping network training flowchart of an embodiment of the present application;

[0047] Figure 5 A face recognition network training flowchart of an embodiment of the present application. DETAILED DESCRIPTION

[0048] It should be noted that:

[0049] Since accurate recognition is also very important for face recognition in an environment with abnormal lighting conditions, in order to improve the accuracy of face recognition in an environment with abnormal lighting conditions, exposure image correction needs to be performed to improve the face recognition effect. Therefore, the present application proposes a face recognition method and system based on an image exposure correction network to solve the problem of low accuracy of the current face recognition model in an environment with abnormal lighting conditions, and to enhance the accuracy of the face recognition model.

[0050] To make the objectives, technical solutions and advantages of embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application. In addition, the technical features in each embodiment or single embodiment provided by the present application can be combined with each other at will to form new technical solutions. Such combination is not subject to the order of steps and / or structure composition mode, but must be based on the fact that it can be realized by those of ordinary skill in the art. When the combination of technical solutions appears to be contradictory or unachievable, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present application.

[0051] The present embodiment takes a given face recognition data set as an example to further illustrate the present application. Please see Figure 1 The face recognition method based on the image exposure correction network provided by the present embodiment comprises the following steps:

[0052] Step 1, pre-processing the input image;

[0053] Step 2, inputting the pre-processed image into a face image extraction network to obtain a face image;

[0054] Step 3, inputting the face image into an image exposure correction network to obtain a normal exposure face image;

[0055] Step 4, inputting the normal exposure face image into a face recognition network to obtain a face recognition result.

[0056] The face image extraction network comprises a HOG feature extraction module, an input of which is a to-be-detected image, and an output of which is a face image.

[0057] See Figure 2 The image exposure correction network comprises a color space conversion module, an exposure feature extraction network, a brightness mapping network, and a detail enhancement module.

[0058] In an embodiment, the color space conversion module has an input of a face RGB image and an output of a face HSV image.

[0059] The exposure feature extraction network comprises a down-sampling layer, a random cropping module, and an image encoder A module, the input dimension, the input channel number, and the output channel number of the entire module are the same as those of the input image, and the output dimension is a hyperparameter in the design.

[0060] The down-sampling layer comprises four basic network structures and one sampling layer, each basic network structure comprises two residual blocks and one self-attention layer.

[0061] The image encoder A module comprises a residual neural network and one self-attention layer.

[0062] The brightness mapping network has an input of an overexposed or underexposed image feature representation and an output of an exposure-normalized image.

[0063] The detail enhancement module has an input of a detail-lacking image and an output of a detail-enhanced image.

[0064] The face recognition network is composed of a SimCLR-based image encoder B, and outputs a face recognition result.

[0065] The SimCLR-based image encoder B comprises two residual blocks and one self-attention layer.

[0066] The image exposure correction network generates an exposure-normalized face image, which is input into the face recognition network, and outputs a face recognition result.

[0067] In an embodiment, in step 1, the input image is preprocessed, including normalization of the input image, the normalization uses scale normalization to scale all sample data to the same scale, and is symmetrically distributed about 0, and the specific normalization function is:

[0068]

[0069] wherein y is the normalized image, and x is the input noise image.

[0070] In one implementation, in step 2, the facial image extraction network includes HOG feature extraction. The principle of HOG extraction is to first convert the image to grayscale, then calculate the gradient of the pixels in the image, and by converting the image into HOG form, the position of the face in the image can be obtained.

[0071] In one implementation, in step 3, the color space conversion module converts the input RGB face image into an HSV image, using the following specific calculation formula:

[0072]

[0073] In one implementation, in step 3, the exposure feature extraction network has a downsampling layer consisting of four basic network structures and one sampling function. The basic network structures include two residual blocks and one self-attention layer. The downsampling function consists of a convolution and a Rearrange function, performing downsampling through dimension merging. Each residual block consists of two convolutional layers, and the residual block computation unit can be represented as:

[0074]

[0075] Where y is the output of the residual block. is the output of the second convolutional layer in the residual block, and x is the input of the residual block. The residual structure is used so that the network can "short-circuit" some layers when it is deep to ensure the effective transmission of information.

[0076] In one implementation, in step 3, the exposure feature extraction network incorporates a self-attention module in its downsampling layer. This self-attention module contains three weight matrices with optimized parameters. It is used to perform matrix multiplication with the input image to obtain a three-parameter matrix. The specific calculation formula is as follows:

[0077]

[0078] Where x represents the output feature of the convolutional layer in the residual neural network;

[0079] Therefore, the formula for calculating the weight matrix of the entire image is:

[0080]

[0081] in, for variance The normalization function is defined by the following formula:

[0082]

[0083] in, to find the maximum of the vector .

[0084] See Figure 3 In an embodiment, in step 3, the exposure feature extraction network, the image encoder A module is composed of a residual neural network and one self-attention layer. The training process of the exposure feature extraction network includes the following sub-steps:

[0085] Step S1, import the exposure error image dataset;

[0086] The exposure error image dataset uses Exposure-Errors Dataset and DEPD;

[0087] Step S2, input the exposure error image and the exposure normal image into the same down-sampling layer respectively;

[0088] The exposure error image and the exposure normal image are input into the same down-sampling layer respectively, and the structure of the down-sampling layer is the same as that described in step 3, to ensure the consistency of the training environment and the application environment;

[0089] Step S3, input the down-sampled exposure error image and the exposure normal image and the original exposure error image and the exposure normal image into the random cropping module respectively, which provides images with fixed resolution and enriches the diversity of training data;

[0090] Step S4, input the randomly cropped image into the image encoder A module, and then project the output of the encoder A into the feature space through the decoder, wherein the decoder is a multi-layer perception (MLP) which reduces the dimension of the representation generated by the encoder, and the process can be represented as:

[0091]

[0092] wherein x represents the input image, and respectively represent the encoder A network and the decoder network, h is the k-dimensional data output by the encoder (k is a hyperparameter in the design), and z is the feature vector of the output of the encoder A projected into the feature space through the decoder;

[0093] Step S5, input the training data into the image encoder A and the decoder network for training, and constantly optimize the model parameters through the normalized temperature scaled cross-entropy loss function;

[0094] wherein the expression of the normalized temperature scaled cross-entropy loss function is:

[0095]

[0096] where N is the number of images in the batch, is the indicator function, is the temperature parameter, to measure the similarity between a and b.

[0097] See Figure 4 In an embodiment, in step 3, the image exposure correction network, the brightness mapping network input is the feature representation of the output of the exposure feature extraction network described above, and the output is the exposure-corrected image data, the training process of the brightness mapping network includes the following sub-steps:

[0098] Step S1, import the exposure error image dataset;

[0099] The exposure error image dataset uses Exposure-Errors Dataset and DEPD;

[0100] Step S2, train the exposure error image and the exposure normal image on the trained frozen exposure feature extraction network using the regularized linear regressor (ridge regression), to obtain the brightness mapping relationship.

[0101] In an embodiment, in step 3, the image exposure correction network, the detail enhancement module uses a Gaussian filter to perform smoothing filtering on the brightness-mapped image to obtain a detail-enhanced image.

[0102] See Figure 5 In an embodiment, in step 4, the face recognition network is composed of an image encoder B based on SimCLR, and the training process of the face recognition network includes the following sub-steps:

[0103] Step S1, import the face recognition dataset;

[0104] The face recognition dataset uses the Glint360K dataset;

[0105] Step S2, randomly perform image enhancement on the face image;

[0106] Randomly performing image enhancement on the face image includes rotating the image, changing the color of the image, changing the saturation of the image, changing the brightness of the image, scaling, cropping, etc. Two image enhancement methods are randomly used to enhance the image data.

[0107] Step S3, input the face image into the image encoder B trained based on SimCLR, and continuously optimize the model parameters through gradient descent and back propagation to obtain a contrast pre-training module;

[0108] Wherein SimCLR is an unsupervised learning method for learning the representation of images, the core idea is to train the neural network through contrastive learning, let the model learn to distinguish whether the views obtained by image enhancement from multiple images are from the same original image.

[0109] Wherein the image encoder B is composed of one basic network structure, the basic network structure includes two residual blocks and one self-attention layer, and each residual block is composed of two convolutional layers.

[0110] The implementation basis of each embodiment of the present application is realized by the programmed processing of the device with the processor function. Therefore, in engineering practice, the technical scheme and function of each embodiment of the present application are packaged into various modules. Based on this actual situation, on the basis of each embodiment described above, the embodiment of the present application provides a face recognition system based on image exposure correction network, which is used to execute the face recognition method based on image exposure correction network in the method embodiment.

[0111] The system comprises: a preprocessing unit configured to preprocess an input image to obtain a normalized image; a face image extraction unit configured to input the preprocessed image into a face image extraction network to obtain a face image; and an image exposure correction unit configured to input the face image into an image exposure correction network to obtain an image with normal exposure. The image exposure correction network comprises a color space conversion module, an exposure feature extraction network, a brightness mapping network and a detail enhancement module. The color space conversion module takes a face RGB image as input and outputs a face HSV image. The exposure feature extraction network comprises a down-sampling layer, a random cropping module and an image encoder module. The input dimension, input channel number and output channel number of the entire module are the same as those of the input image, and the output dimension is a hyperparameter in the design. The down-sampling layer comprises four basic network structures and one sampling layer. Each basic network structure comprises two residual blocks and one self-attention layer. The image encoder A module comprises a residual neural network and one self-attention layer. The brightness mapping network takes an overexposed or underexposed image feature representation as input and outputs an image with normal exposure. The detail enhancement module takes a detail-lacking image as input and outputs a detail-enhanced image. A face recognition unit is configured to input the image with normal exposure into a SimCLR-based image encoder B and output a face recognition result. The SimCLR-based image encoder B comprises two residual blocks and one self-attention layer. The image exposure correction network generates a face image with normal exposure and inputs it into a face recognition module to output a face recognition result.

[0112] The specific implementation modes of each unit are the same as those of each step, and the present application will not be described.

[0113] The face recognition system based on the image exposure correction network provided by the embodiment of the present application can greatly improve the accuracy of face recognition under the condition of abnormal illumination environment.

[0114] It should be noted that the system embodiments provided by the present application are used to implement the methods in the above method embodiments, and are also used to implement the methods in other method embodiments provided by the present application. The difference is only that the corresponding function modules are set, and the principle is basically the same as that of the above system embodiments provided by the present application. As long as the person skilled in the art improves the modules in the above system embodiments on the basis of the above system embodiments, refers to the specific technical solutions in other method embodiments, obtains the corresponding technical means by combining technical features, and the technical solutions composed of these technical means, on the premise of ensuring the practicability of the technical solutions, the corresponding system class embodiments are obtained, which are used to implement the methods in other method class embodiments.

[0115] Based on the same inventive concept as the above embodiments, the embodiment of the present application also provides a face recognition device based on an image exposure correction network, comprising:

[0116] One or more processors;

[0117] A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement a face recognition method based on an image exposure correction network as claimed in any one of claims 1 to 10.

[0118] In summary of the above embodiments, the present application can realize face recognition under the condition of abnormal illumination environment. Compared with the traditional face recognition model, the present application has the advantages of stability, independence, rapidness and efficiency, can greatly improve the accuracy of face recognition under the condition of abnormal illumination environment, and has good popularization and application prospect.

[0119] The terms "include" and "have" and any variations thereof in the specification and claims of the present application and the above drawings are intended to cover the non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to the steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0120] It should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present application.

Claims

1. A face recognition method based on an image exposure correction network, characterized in that, include: The image to be detected is input into a facial image extraction network to obtain a face image; the facial image extraction network includes a HOG feature extraction module. The obtained face image is input into an image exposure correction network to obtain a properly exposed face image. The image exposure correction network includes a color space conversion module, an exposure feature extraction network, a brightness mapping network, and a detail enhancement module. The input of the exposure feature extraction network is an overexposed or underexposed image, and the output is an overexposed or underexposed image feature representation. The input of the brightness mapping network is an overexposed or underexposed image feature representation, and the output is a properly exposed image. The input of the detail enhancement module is a properly exposed image with missing details, and the output is a detail-enhanced image. The exposure feature extraction network includes a downsampling layer, a random cropping module, and an image encoder module. The downsampling layer contains four basic network structures and one sampling layer. Each basic network structure includes two residual blocks and one self-attention layer. The random cropping module is used to provide an image with a fixed resolution. The image encoder module includes a residual neural network and a self-attention layer; the training of the exposure feature extraction network includes: importing an exposure error image dataset; inputting the exposure error images and the exposure normal images into the same downsampling layer; inputting the downsampled exposure error images and the exposure normal images, as well as the original exposure error images and the exposure normal images, into a random cropping module; inputting the randomly cropped images into the image encoder module, and then projecting the output of the image encoder module onto the feature space through a decoder; inputting training data into the image encoder module and the decoder network for training, and continuously optimizing the model parameters through a standardized temperature-scaled cross-entropy loss function, outputting the trained exposure feature extraction network; The obtained normally exposed face image is input into the face recognition network to obtain the face recognition result; the face recognition network includes an image encoder based on SimCLR.

2. The face recognition method based on an image exposure correction network according to claim 1, characterized in that, The method further includes: The image to be detected is acquired and preprocessed.

3. The face recognition method based on an image exposure correction network according to claim 1, characterized in that, The training of the brightness mapping network includes: Import the dataset of images with exposure errors; The brightness mapping relationship is obtained by training a regularized linear regressor on a pre-trained frozen exposure feature extraction network using images with incorrect exposure and images with correct exposure.

4. The face recognition method based on an image exposure correction network according to claim 1, characterized in that, The implementation of the detail enhancement module includes: A Gaussian filter is used to smooth the brightness-mapped image, resulting in an image with enhanced details.

5. The face recognition method based on an image exposure correction network according to claim 1, characterized in that, The training of the face recognition network includes: Import the face recognition dataset; Random image enhancement is applied to facial images; The enhanced face image is input into the image encoder trained based on SimCLR. The model parameters are continuously optimized through gradient descent and backpropagation to obtain the comparison pre-trained module. The image encoder includes a basic network structure, which includes two residual blocks and one self-attention layer. Each residual block consists of two convolutional layers.

6. A face recognition system based on an image exposure correction network, characterized in that, include: A facial image extraction unit is used to input the image to be detected into a facial image extraction network to obtain a face image; the facial image extraction network includes a HOG feature extraction module. An image exposure correction unit is used to input the obtained face image into an image exposure correction network to obtain a properly exposed face image. The image exposure correction network includes a color space conversion module, an exposure feature extraction network, a brightness mapping network, and a detail enhancement module. The input to the exposure feature extraction network is an overexposed or underexposed image, and the output is an overexposed or underexposed image feature representation. The input to the brightness mapping network is an overexposed or underexposed image feature representation, and the output is a properly exposed image. The input to the detail enhancement module is a properly exposed image with missing details, and the output is a detail-enhanced image. The exposure feature extraction network includes a downsampling layer, a random cropping module, and an image encoder module. The downsampling layer contains four basic network structures and one sampling layer; each basic network structure includes two residual blocks and one self-attention layer. The random cropping module is used to provide an image at a fixed resolution. The image encoder module includes a residual neural network and a self-attention layer; the training of the exposure feature extraction network includes: importing an exposure error image dataset; inputting the exposure error images and the exposure normal images into the same downsampling layer; inputting the downsampled exposure error images and the exposure normal images, as well as the original exposure error images and the exposure normal images, into a random cropping module; inputting the randomly cropped images into the image encoder module, and then projecting the output of the image encoder module onto the feature space through a decoder; inputting training data into the image encoder module and the decoder network for training, and continuously optimizing the model parameters through a standardized temperature-scaled cross-entropy loss function, outputting the trained exposure feature extraction network; A face recognition unit is used to input a normally exposed face image into a face recognition network to obtain a face recognition result; the face recognition network includes an image encoder based on SimCLR.

7. A face recognition device based on an image exposure correction network, characterized in that, include: One or more processors; And a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the steps of a face recognition method based on an image exposure correction network as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the steps of a face recognition method based on an image exposure correction network as described in any one of claims 1 to 5.

Images