Dynamic threshold face recognition method and device based on face quality

Abstract

The invention discloses a dynamic threshold face recognition method and device based on face quality, and relates to the technical field of face recognition. The method comprises the following steps: a model construction training process: constructing a face quality evaluation model and a training sample, the training sample comprising a face image quality score label, the face image quality score being used for reflecting the definition of a face image, light conditions and the recognizable degree of facial features, the higher the face image quality score is, the better the face quality is; training a face quality evaluation model by using the training sample; a dynamic threshold face recognition process: inputting a to-be-recognized face image into the trained face quality evaluation model to obtain a corresponding face image quality score; the comparison threshold value is calculated through the piecewise function according to the obtained face image quality score, then face recognition is carried out according to the adjusted comparison threshold value, and the accuracy of face recognition and the adaptability to various environmental conditions are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and in particular to a dynamic threshold face recognition method and device based on face quality. Background Art

[0002] Face recognition technology is a biometric technology that identifies or verifies an individual's identity by analyzing the visual features of the face. This technology has a wide range of applications in security monitoring, access control, identity verification, and other fields.

[0003] Although existing face recognition technology has made significant progress, there are still some limitations and problems: 1. Fixed comparison threshold: Many existing face recognition systems use a fixed comparison threshold to decide whether to accept or reject the recognition result. This method cannot adapt to images of different quality or conditions, and may result in long-term failure in complex scenes, such as low-quality images or changing lighting conditions.

[0004] These problems highlight the limitations of face recognition technology in complex environments and conditions. Existing face recognition technology needs further development and innovation to improve its performance and reliability in various environments and conditions. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a dynamic threshold face recognition method and device based on face quality, which improves the accuracy of face recognition and adaptability to various environmental conditions by carefully evaluating the quality of face images and then dynamically adjusting the comparison threshold according to the evaluation results.

[0006] In a first aspect, the present invention provides a dynamic threshold face recognition method based on face quality, comprising:

[0007] Model building and training process: construct a face quality assessment model and training samples, wherein the training samples include a face image quality score label. The face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the degree of recognizability of facial features. The higher the face image quality score, the better the face quality. Then use the training samples to train the face quality assessment model.

[0008] Dynamic threshold face recognition process: input the face image to be recognized into the trained face quality assessment model to obtain the corresponding face image quality score; calculate the comparison threshold through the piecewise function according to the obtained face image quality score, and then perform face recognition according to the adjusted comparison threshold. The formula of the piecewise function is:

[0009]

[0010] Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

[0011] In a second aspect, the present invention provides a dynamic threshold face recognition device based on face quality, comprising:

[0012] A model building training module is used to build a face quality assessment model and training samples, wherein the training samples include a face image quality score label, and the face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the degree of recognizability of facial features. The higher the face image quality score, the better the face quality. The training samples are then used to train the face quality assessment model.

[0013] The dynamic threshold face recognition module is used to input the face image to be recognized into the trained face quality assessment model to obtain the corresponding face image quality score; calculate the comparison threshold through the piecewise function according to the obtained face image quality score, and then perform face recognition according to the adjusted comparison threshold. The formula of the piecewise function is:

[0014]

[0015] Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

[0016] The technical solution provided in the embodiments of the present invention has at least the following technical effects:

[0017] The dynamic threshold mechanism based on face quality enables the face recognition system to automatically adjust the comparison threshold according to the image quality, thereby showing higher adaptability in different environments and scenarios; this adaptive capability makes the face recognition system more flexible and reliable in practical applications, thereby improving the overall performance, especially when dealing with low-quality images or complex lighting conditions.

[0018] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0020] Figure 1 This is a schematic diagram of the process of the method in Embodiment 1 of the present invention;

[0021] Figure 2 This is a schematic diagram of the process of dynamic threshold face recognition in the first embodiment of the present invention;

[0022] Figure 3 It is a schematic diagram of the structure of the device in the second embodiment of the present invention. DETAILED DESCRIPTION

[0023] The embodiments of the present invention provide a dynamic threshold face recognition method and device based on face quality, which performs a detailed evaluation on the quality of face images and then dynamically adjusts the comparison threshold according to the evaluation results, thereby improving the accuracy of face recognition and the adaptability to various environmental conditions.

[0024] The technical solution in the embodiment of the present invention has the following general idea:

[0025] By proposing an innovative face recognition method, the core of which is to finely evaluate the quality of face images to achieve the purpose of dynamically adjusting the comparison threshold. The design concept of the present invention aims to improve the accuracy of face recognition and the adaptability to various environmental conditions, especially in changing and challenging scenes.

[0026] The main technical improvements of the embodiments of the present invention are concentrated in the following aspects:

[0027] 1. Developing a face quality assessment model: An important improvement in the embodiment of the present invention is the development of a face quality assessment model based on a lightweight convolutional neural network, which can evaluate the quality of input face images, reflecting the image clarity, lighting conditions, and the degree of recognizability of facial features. Through this model, the quality of face images can be quantitatively analyzed, providing key support for the performance optimization of face recognition systems.

[0028] 2. Dynamic threshold mechanism based on face quality: In current face recognition technology, the face comparison process usually uses a fixed comparison threshold, which will lead to a decrease in recognition rate when processing low-quality images such as blur and occlusion. The embodiment of the present invention uses the face image quality assessment model described in point 1 to assess the quality of face images, and dynamically adjusts the comparison threshold based on the assessment results. For images with lower quality, the comparison threshold will be appropriately lowered to improve the recognition accuracy. This method can better adapt to complex and changing environmental conditions, thereby improving the overall face recognition performance.

[0029] The core innovation of the embodiment of the present invention lies in the dynamic threshold adjustment mechanism based on face quality. The introduction of this mechanism realizes a face recognition method with higher recognition rate and stronger adaptability, which can maintain a high level of recognition performance in various environments and conditions.

[0030] Embodiment 1

[0031] This embodiment provides a dynamic threshold face recognition method based on face quality, such as Figure 1 As shown, including:

[0032] S1. Model building and training process: construct a face quality assessment model and training samples, wherein the training samples include a face image quality score label. The face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the recognizability of facial features. The higher the face image quality score, the better the face quality. Then use the training samples to train the face quality assessment model.

[0033] The criteria for judging the quality score of a face image can be based on a variety of factors, such as:

[0034] The higher the image clarity, the higher the face image quality score. If the image is blurry, the face image quality score is low.

[0035] The better the lighting conditions, the higher the face image quality score. If the lighting conditions are bad, such as in a dark scene, the face image quality score is low.

[0036] The more complete the facial features are, the higher the face image quality score is. If the facial features are incomplete, such as being occluded, the face image quality score is low.

[0037] A convolutional neural network can be trained with a large number of face images to create a face quality assessment model that can analyze the quality of face images and help improve the accuracy of face recognition.

[0038] S2, dynamic threshold face recognition process: Figure 2 As shown, the following steps may be specifically included:

[0039] S21, face image quality assessment process: the face image to be recognized is aligned and then input into the trained face quality assessment model to obtain the corresponding face image quality score;

[0040] S22, dynamic adjustment process of comparison threshold: the comparison threshold T is calculated by piecewise function according to the obtained face image quality score, and then face recognition is performed according to the adjusted comparison threshold T. The formula of the piecewise function is:

[0041]

[0042] Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

[0043] Specifically, in the formula of the piecewise function, the range of score is between 0 and 1, the preset value of t1 is 0.5, the preset value of t2 is 0.07, the preset value of min is 0.3, and the preset value of max is 0.6.

[0044] In one possible implementation, a large number of face images are used to train the face quality assessment model through knowledge distillation. Knowledge distillation is a technique that compresses the knowledge of the teacher model and transfers it to the student model. Specifically, the teacher model first predicts the input data, and then the prediction results of the teacher model are used as training labels for the student model. In this way, the student model can acquire the knowledge of the teacher model, so that the performance is close to or even better than the teacher model, while keeping the model size and computational cost small.

[0045] 1) Teacher model: This model has usually been trained on a large-scale dataset and has high accuracy and strong learning ability. It is a large model.

[0046] 2) Student model: This model has a relatively simple structure, a small number of parameters, and high computational efficiency. It is a small model.

[0047] In this embodiment, the teacher model uses the CR-FIQA model, which is an open source face quality assessment model trained based on the ResNet-100 model. The student model uses the MobileneV3 model. MobileNetV3 is an efficient and lightweight neural network designed specifically for running on mobile and edge devices. It can significantly reduce the consumption of computing resources while maintaining a high recognition accuracy. During the training process, SmoothL1 loss is used as the training loss function to optimize the network parameters. The formula is as follows:

[0048]

[0049] Among them, y represents the prediction score of the teacher model, and y' represents the prediction score of the student model.

[0050] Through knowledge distillation, the student model can learn how to identify and quantify various factors that affect the quality of facial images, such as clarity, lighting conditions, and the recognizability of facial features, from the teacher model. Ultimately, the trained student model is used as a face quality assessment model for subsequent use, which can provide a key quality assessment reference for the subsequent face recognition process.

[0051] The face recognition method of this embodiment was applied to a test experiment of about 12,000 face images taken in various environments including dark light, strong light, and blur. The results showed that when TAR@FAR=1e-6, the TAR of the face recognition system was improved from 89.527% to 94.511%.

[0052] Based on the same inventive concept, the present application also provides a device corresponding to the method in Example 1, see Example 2 for details.

[0053] Embodiment 2

[0054] In this embodiment, a dynamic threshold face recognition device based on face quality is provided. Figure 3 As shown, including:

[0055] A model building training module is used to build a face quality assessment model and training samples, wherein the training samples include a face image quality score label, and the face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the degree of recognizability of facial features. The higher the face image quality score, the better the face quality. The training samples are then used to train the face quality assessment model.

[0056] The dynamic threshold face recognition module is used to input the face image to be recognized into the trained face quality assessment model to obtain the corresponding face image quality score; calculate the comparison threshold through the piecewise function according to the obtained face image quality score, and then perform face recognition according to the adjusted comparison threshold. The formula of the piecewise function is:

[0057]

[0058] Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

[0059] Specifically, in the formula of the piecewise function, the range of score is between 0 and 1, the preset value of t1 is 0.5, the preset value of t2 is 0.07, the preset value of min is 0.3, and the preset value of max is 0.6.

[0060] Preferably, in the model construction training module, the face quality assessment model is trained by knowledge distillation, the teacher model first predicts the input training samples, and then the prediction results of the teacher model are used as training labels for the student model, and the trained student model is used as the trained face quality assessment model;

[0061] Preferably, in the model construction training module, the teacher model is the CR-FIQA model and the student model is the MobileNetV3 model.

[0062] Preferably, in the model building training module, the model building training process uses Smooth L1 loss as the training loss function, and the formula is as follows:

[0063]

[0064] Among them, y represents the prediction score of the teacher model, and y' represents the prediction score of the student model.

[0065] Since the device introduced in the second embodiment of the present invention is a device used to implement the method of the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the device based on the method introduced in the first embodiment of the present invention, so it is not described here in detail. All devices used in the method of the first embodiment of the present invention belong to the scope of protection of the present invention.

[0066] The dynamic threshold mechanism based on face quality of the present invention enables the face recognition system to automatically adjust the comparison threshold according to the image quality, thereby showing higher adaptability in different environments and scenarios; this adaptive capability makes the face recognition system more flexible and reliable in practical applications, thereby improving the overall performance, especially when processing low-quality images or complex lighting conditions.

[0067] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0071] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A dynamic threshold face recognition method based on face quality, characterized in that: include: Model building and training process: construct a face quality assessment model and training samples, wherein the training samples include a face image quality score label. The face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the degree of recognizability of facial features. The higher the face image quality score, the better the face quality. Then use the training samples to train the face quality assessment model. Dynamic threshold face recognition process: input the face image to be recognized into the trained face quality assessment model to obtain the corresponding face image quality score; calculate the comparison threshold through the piecewise function according to the obtained face image quality score, and then perform face recognition according to the adjusted comparison threshold. The formula of the piecewise function is: Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

2. The method according to claim 1, characterized in that: In the formula of the piecewise function, the range of score is between 0 and 1, the preset value of t1 is 0.5, the preset value of t2 is 0.07, the preset value of min is 0.3, and the preset value of max is 0.

6.

3. The method according to claim 1, characterized in that: During the model construction and training process, the face quality assessment model is trained by knowledge distillation. The teacher model first predicts the input training samples, and then the prediction results of the teacher model are used as training labels for the student model. The trained student model is used as the trained face quality assessment model.

4. The method according to claim 3, characterized in that: The teacher model is the CR-FIQA model, and the student model is the MobileNetV3 model.

5. The method according to claim 3, characterized in that: The model construction and training process uses Smooth L1 loss as the training loss function, and the formula is as follows: Among them, y represents the prediction score of the teacher model, and y' represents the prediction score of the student model.

6. A dynamic threshold face recognition device based on face quality, characterized in that: include: A model building training module is used to build a face quality assessment model and training samples, wherein the training samples include a face image quality score label, and the face image quality score is used to reflect the clarity of the face image, the lighting conditions, and the degree of recognizability of facial features. The higher the face image quality score, the better the face quality. The training samples are then used to train the face quality assessment model. The dynamic threshold face recognition module is used to input the face image to be recognized into the trained face quality assessment model to obtain the corresponding face image quality score; calculate the comparison threshold through the piecewise function according to the obtained face image quality score, and then perform face recognition according to the adjusted comparison threshold. The formula of the piecewise function is: Among them, t1 is the benchmark comparison threshold used to determine whether it is the same identity; t2 is the descending range of the threshold adjustment, which is used to constrain the adjustment range of the comparison threshold T; score is the face image quality score; min and max represent the minimum and maximum values ​​of the boundary points of the piecewise function, respectively.

7. The device according to claim 6, characterized in that: In the formula of the piecewise function, the range of score is between 0 and 1, the preset value of t1 is 0.5, the preset value of t2 is 0.07, the preset value of min is 0.3, and the preset value of max is 0.

6.

8. The device according to claim 6, characterized in that: In the model construction training module, the face quality assessment model is trained by knowledge distillation. The teacher model first predicts the input training samples, and then the prediction results of the teacher model are used as training labels for the student model. The trained student model is used as the trained face quality assessment model.

9. The device according to claim 8, characterized in that: In the model construction training module, the teacher model is the CR-FIQA model and the student model is the MobileNetV3 model.

10. The device according to claim 8, characterized in that: In the model building training module, the model building training process uses Smooth L1 loss as the training loss function, and the formula is as follows: Among them, y represents the prediction score of the teacher model, and y' represents the prediction score of the student model.

Images