Face recognition method and system based on image analysis

Abstract

The invention provides a face recognition method and system based on image analysis, and relates to the technical field of image analysis, and the method comprises the steps: determining inter-frame difference vectors through the difference information between adjacent frames in continuous image frames during face recognition, carrying out the posture fusion of the continuous image frames through all inter-frame difference vectors, and obtaining a posture fusion feature; mapping the attitude fusion feature to a face contour recognition space based on a space difference angle during face recognition, performing deconvolution on the attitude fusion feature in the face contour recognition space, and determining an attitude mapping cost during face recognition according to a local contour feature obtained by deconvolution; performing recognition compensation on a face recognition process according to the environment illumination intensity and the mapping cost to obtain a compensation characteristic quantity; according to the face recognition method and device, the dynamic feature adjustment and light compensation in the face recognition process can be realized, and the accuracy of face recognition is improved.

Description

Technical Field

[0001] This application relates to the field of image analysis technology. More specifically, this application relates to a face recognition method and system based on image analysis. Background Art

[0002] Image analysis is a technology based on computer vision and digital image processing, aiming to automatically extract, identify, and analyze key information in images through algorithms. The core technologies of image analysis include edge detection, feature extraction, pattern recognition, deep learning, etc., and can be used in multiple fields such as object recognition, target detection, and scene understanding. Among them, in the field of face recognition, image analysis technology plays a crucial role.

[0003] Face recognition is an identity verification method based on biometric recognition technology, which determines a person's identity by analyzing and comparing personal facial features. The main goal is to enable a computer to recognize and verify faces like humans. Face recognition technology is widely used in fields such as security monitoring and access control systems, and has received extensive attention and use due to its contactless nature and high accuracy. Traditional face recognition methods rely on key point detection and template matching. Due to the real-time changes in face poses, which can cause the loss of facial pose features, and different environmental light intensities, which can cause uneven facial brightness, etc., it is difficult for existing face matching algorithms to effectively match facial features. Therefore, how to achieve dynamic feature adjustment and light compensation in the face recognition process to improve the accuracy of face recognition is a difficult problem faced by the current industry. Summary of the Invention

[0004] This application provides a face recognition method and system based on image analysis, which can achieve dynamic feature adjustment and light compensation in the face recognition process to improve the accuracy of face recognition.

[0005] In a first aspect, this application provides a face recognition method based on image analysis. The face recognition method includes: Collect consecutive image frames during face recognition; Determine an inter-frame difference vector based on the difference information between adjacent frames in the consecutive image frames, and perform pose fusion on the consecutive image frames through all the inter-frame difference vectors to obtain a pose fusion feature; Map the pose fusion feature to a facial contour recognition space based on the spatial difference angle during face recognition, perform deconvolution on the pose fusion feature in the facial contour recognition space, and then determine the pose mapping cost during face recognition from the local contour features obtained by the deconvolution; Obtain the environmental light intensity during face recognition, and perform recognition compensation on the face recognition process based on the environmental light intensity and the mapping cost to obtain a compensation feature quantity; Determine the matching confidence of face recognition through the compensation feature quantity, and then screen the face recognition results according to the matching confidence.

[0006] In this embodiment, continuous image frames during face recognition are collected by a depth camera.

[0007] In this embodiment, determining the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames specifically includes: Extract the key feature maps in each image frame from the continuous image frames; Determine the difference images of adjacent frames in the continuous image frames, and then determine the inter-frame difference vector of the adjacent frames from the difference images of the adjacent frames and all the key feature maps.

[0008] In this embodiment, performing pose fusion on the continuous image frames through all the inter-frame difference vectors to obtain pose fusion features specifically includes: Construct a pose fusion matrix based on all the inter-frame difference vectors; Determine the pose fusion features through the eigenvalues of the pose fusion matrix.

[0009] In this embodiment, mapping the pose fusion features to the facial contour recognition space based on the spatial difference angle during face recognition specifically includes: Determine the mapping angle according to the spatial difference angle; Perform a linear contour mapping on the pose fusion features based on the mapping angle to obtain the facial contour recognition space.

[0010] In this embodiment, determining the pose mapping cost during face recognition from the local contour features obtained by deconvolution specifically includes: Determine the facial pose fusion map according to the pose fusion features in the facial contour recognition space; Determine the pose loss vector during face recognition from the facial pose fusion map and all the local contour features; Determine the pose mapping cost during face recognition through the convolution scale of deconvolution of the pose loss vector during face recognition and the pose fusion features in the facial contour recognition space.

[0011] In this embodiment, perform deconvolution on the pose fusion features in the facial contour recognition space through a preset neural network model.

[0012] In this embodiment, obtain the ambient light intensity during face recognition through the light sensor of the depth camera.

[0013] In this embodiment, performing recognition compensation on the face recognition process according to the ambient light intensity and the mapping cost to obtain the compensation feature quantity specifically includes: Determine the feature correction amount in the face recognition process according to the mapping cost; Determine the light compensation coefficient according to the environmental light intensity; Obtain the brightness reference value of the face contour recognition space; Based on the feature correction amount, the light compensation coefficient, and the brightness reference value, respectively perform feature adjustment and light compensation on the face recognition process, and then obtain the compensated feature amount.

[0014] In a second aspect, the present application provides a face recognition system based on image analysis for executing a face recognition method based on image analysis. The face recognition system includes: An image acquisition module for acquiring continuous image frames during face recognition; An attitude fusion module for determining the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames, and performing attitude fusion on the continuous image frames through all the inter-frame difference vectors to obtain the attitude fusion feature; A feature processing module for mapping the attitude fusion feature to the face contour recognition space based on the spatial difference angle during face recognition, performing deconvolution on the attitude fusion feature in the face contour recognition space, and then determining the attitude mapping cost during face recognition from the local contour features obtained by the deconvolution; A light compensation module for obtaining the environmental light intensity during face recognition, and performing recognition compensation on the face recognition process according to the environmental light intensity and the mapping cost to obtain the compensated feature amount; A matching and screening module for determining the matching confidence of face recognition through the compensated feature amount, and then screening the face recognition results according to the matching confidence.

[0015] The technical solution provided by the disclosed embodiments of the present application has the following beneficial effects: By acquiring continuous image frames during face recognition; determining the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames, performing attitude fusion on the continuous image frames through all the inter-frame difference vectors to obtain the attitude fusion feature; mapping the attitude fusion feature to the face contour recognition space based on the spatial difference angle during face recognition, performing deconvolution on the attitude fusion feature in the face contour recognition space, and then determining the attitude mapping cost during face recognition from the local contour features obtained by the deconvolution; obtaining the environmental light intensity during face recognition, performing recognition compensation on the face recognition process according to the environmental light intensity and the mapping cost to obtain the compensated feature amount; determining the matching confidence of face recognition through the compensated feature amount, and then screening the face recognition results according to the matching confidence.

[0016] It can be seen that in this application, dynamic feature adjustment and light compensation can be achieved during the face recognition process. First, by acquiring consecutive image frames, the dynamic features of the face can be captured, reducing recognition errors caused by instantaneous changes, providing more image feature information, and reflecting the change trend of the face in a short period of time through the inter-frame difference vector, facilitating dynamic correction of pose errors, reducing feature deviations caused by head rotation or tilt, making feature extraction more accurate, and improving face matching accuracy under different poses. Second, by adjusting the pose fusion features through the spatial difference angle and mapping them to the standardized facial contour recognition space, the influence caused by individual differences or changes in shooting angles can be reduced. Using the deconvolution operation can restore local contour details, making the facial features clearer and conducive to improving the accuracy of feature matching. Then, by obtaining the ambient light information in real time, the images in low-light or strong-light conditions can be dynamically adjusted, and compensation can be combined with the pose mapping cost, effectively reducing the influence of light changes on feature extraction and improving the adaptability of the system to different light conditions. Finally, by compensating the feature quantity, the final matching result of face recognition can be optimized, improving the adaptability of the system to dynamic changes, reducing recognition errors, and improving the accuracy of face recognition.

[0017] In summary, the technical solution adopted in this application can achieve dynamic feature adjustment and light compensation during the face recognition process to improve the accuracy of face recognition. Brief Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of a face recognition method based on image analysis provided by the present application; Figure 2 It is an exemplary flowchart for determining the pose fusion features provided by the present application; Figure 3 It is an exemplary flowchart for determining the pose mapping cost provided by the present application; Figure 4 It is a module structure diagram of a face recognition system provided by the present application. Detailed Embodiments

[0020] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying 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. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0021] The embodiments of the present application provide a face recognition method and system based on image analysis. The core is to collect continuous image frames during face recognition, determine the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames, and perform pose fusion on the continuous image frames through all the inter-frame difference vectors to obtain pose fusion features; secondly, map the pose fusion features to the facial contour recognition space based on the spatial difference angle during face recognition, perform deconvolution on the pose fusion features in the facial contour recognition space, and then determine the pose mapping cost during face recognition from the local contour features obtained by deconvolution; then, obtain the environmental light intensity during face recognition, and perform recognition compensation on the face recognition process according to the environmental light intensity and the mapping cost to obtain a compensation feature quantity; finally, determine the matching confidence of face recognition through the compensation feature quantity, and then screen the face recognition results according to the matching confidence.

[0022] Embodiment 1. To better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings of the specification and specific implementation manners. Refer to Figure 1 As shown, this figure is an exemplary flowchart of a face recognition method based on image analysis according to the present embodiment of the present application. The face recognition method includes the following steps: In step S1, collect continuous image frames during face recognition.

[0023] Specifically, continuous image frames during face recognition are collected through a depth camera. The continuous image frames refer to the image frames of the face video captured by the depth camera within a specified time period, and the specified time period can be set between 3 - 5 s. It should be noted that the continuous image frames are composed of multiple image frames in chronological order, and each image frame is an individual image frame in the face video captured by the depth camera within the specified time period, and each image frame corresponds to the unique pose feature of the human face.

[0024] In step S2, determine the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames, and perform pose fusion on the continuous image frames through all the inter-frame difference vectors to obtain pose fusion features.

[0025] In this embodiment, determining the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames specifically includes: Extract the key feature maps in each image frame from the continuous image frames; Determine the difference images of adjacent frames in the continuous image frames, and then determine the inter-frame difference vectors of the adjacent frames from the difference images of the adjacent frames and all the key feature maps.

[0026] It should be noted that the key feature maps in this application include: structure center map, contour boundary map, direction feature map, and dynamic expression map. The key features include: nose, ears, eyes, and mouth. Among them, the structure center map is represented by the nose in the human face, representing the geometric center of the human face, pose estimation, and depth information. The ears in the human face can be used as the contour boundary map for side face recognition and head contour matching. The eyes in the human face can be used as the direction feature map for gaze tracking and iris recognition. The mouth in the human face can be used as the dynamic expression map for expression analysis, speech recognition, and lip reading detection. By extracting these key feature maps, the robustness of face recognition at different angles can be improved. It should also be noted that the adjacent frames in this application refer to two adjacent image frames.

[0027] In specific implementation, first, all the key features in all image frames can be extracted by setting the size of the Haar-like input features in the Haar cascade detector (i.e., the size of the key features), and then all the key features in each image frame are combined on an image with a pixel size of 400*400 according to the contour positions of the face. The combined image is used as the key feature map, and thus the key feature maps of all image frames are obtained. Secondly, select a group of adjacent frames, convolve and fuse the group of adjacent frames through a convolutional neural network to obtain a difference image, calculate the Euclidean distance between the two image frames in the group of adjacent frames and the difference image, and calculate the rotation angle between the two image frames in the group of adjacent frames and the difference image (i.e., the arctangent value of the ratio of the difference in the vertical coordinates to the difference in the horizontal coordinates of the same pixel point of the two key feature maps). Among them, the calculation process can be provided by the Pandas library. Then, the vector composed of arranging the Euclidean distance and the rotation angle is used as the inter-frame difference vector of the facial pose between the group of adjacent frames, that is, the inter-frame difference vector = (Euclidean distance, rotation angle). Repeat the above steps to obtain the inter-frame difference vectors of the remaining adjacent frames. Among them, the inter-frame difference vector refers to the difference features generated by the pose changes of the key parts of the human face between adjacent frames. By calculating the Euclidean distance and the rotation angle between the key feature maps in all adjacent frames, the changes in the facial poses between the key parts of the human face within a specified time period can be analyzed.

[0028] Preferably, in this embodiment, refer to Figure 2As shown, this figure is an exemplary flowchart for determining pose fusion features in an embodiment of the present application. In this embodiment, pose fusion is performed on the continuous image frames through all inter-frame difference vectors to obtain pose fusion features, which can be specifically implemented by the following steps: First, in step S21, a pose fusion matrix is constructed based on all inter-frame difference vectors; Then, in step S22, the pose fusion features are determined through the eigenvalues of the pose fusion matrix.

[0029] Specifically, in implementation, first, all inter-frame difference vectors can be normalized, and all normalized inter-frame difference vectors are arranged in a column in ascending order according to the sizes of different facial key parts. Then, the matrix formed by sorting and combining all inter-frame difference vectors is used as the pose fusion matrix. The pose fusion matrix reflects the overall pose differences between key feature maps in all adjacent frames. Normalizing all inter-frame difference vectors can eliminate the dimensional differences between different components. Then, all eigenvalues of the pose fusion matrix (i.e., the eigenvalues of the matrix) are obtained through the Pandas library, and the array formed by randomly combining all eigenvalues is used as the pose fusion features.

[0030] It should be noted that the pose fusion features in the present application refer to all eigenvalues under various poses and angles of the human face. The pose fusion features are fused from inter-frame difference vectors generated due to different pose changes of the same recognition target. By determining the pose fusion features, the dynamic changes in the facial pose position and angle can be analyzed, providing strong data support for subsequent pose correction.

[0031] In step S3, based on the spatial difference angle during face recognition, the pose fusion features are mapped to the facial contour recognition space, and deconvolution is performed on the pose fusion features in the facial contour recognition space. Then, the pose mapping cost during face recognition is determined from the local contour features obtained by deconvolution.

[0032] In this embodiment, mapping the pose fusion features to the facial contour recognition space based on the spatial difference angle during face recognition specifically includes: Determining a mapping angle according to the spatial difference angle; Based on the mapping angle, a linear contour mapping is performed on the pose fusion features to obtain the facial contour recognition space.

[0033] It should be noted that the spatial difference angle in the present application refers to the horizontal angle between the position of the webcam and the position of the human face. The spatial difference angle reflects the degree of direction offset during webcam shooting. By using the spatial difference angle during webcam shooting to perform contour mapping on the pose fusion features, the facial contour recognition space of the person can be obtained more accurately.

[0034] In specific implementation, first, obtain the position of the human face and the position of the webcam, determine the connection line between the human face and the webcam, and rotate the spatial difference angle to 0 degrees, so as to use the adjusted angle as the mapping angle; then, initialize a key facial part as the center of the corresponding recognition space. For example, the nose of the human can be initialized as the center of the corresponding recognition space. In other embodiments, other key facial parts can also be set as the center of the recognition space, which is not limited here; secondly, linearly combine all local contour features in the recognition space from large to small; finally, use the recognition space composed of all local contour features after initialization and linear combination as the facial contour recognition space. The facial contour recognition space contains the local contour features corresponding to the combined diagrams of all key facial parts of the human. The local contour features in the facial contour recognition space can more precisely describe the shape and structure of the human face. Different local contour features can represent the difference information of the human face in different poses. Constructing the facial contour recognition space is beneficial to calculating the mapping cost when mapping all local contour features later.

[0035] It should be noted that in this embodiment, the contour mapping of the pose fusion feature is performed through the spatial difference angle, and the facial contour recognition space of the human can be accurately obtained. Considering the different standing positions of the human, the recognition space can be adjusted according to different situations, making the method described in some embodiments of the present application more adaptable.

[0036] In addition, it should also be noted that in the present application, the deconvolution is performed on the pose fusion feature in the facial contour recognition space through a preset neural network model. In specific implementation, the convolution kernel of the transposed convolution is initialized by setting the scale, and the deconvolution is performed through a convolutional neural network with a preset convolution kernel. For example, the scale of the deconvolution is 4, and the size and number of the convolution kernels can be set to 4×4. Among them, the scale in the present application is determined based on the number of key facial parts, that is, when the number of key facial parts in the combined diagram of the key facial parts is set to 4, the scale of the deconvolution is 4. Due to different specific application scenarios, different numbers of key facial parts in the present application can be set according to different application scenarios, which is not limited here.

[0037] Preferably, in this embodiment, refer to Figure 3 As shown, this figure is an exemplary flowchart for determining the pose mapping cost in the embodiment of the present application. In this embodiment, the pose mapping cost during face recognition is determined by the local contour features obtained by deconvolution, and can be specifically implemented by the following steps: First, in step S31, determine the facial pose fusion diagram according to the pose fusion feature in the facial contour recognition space; Then, in step S32, determine the pose loss vector during face recognition through the facial pose fusion diagram and all local contour features; Finally, in step S33, the pose mapping cost during face recognition is determined by deconvolving the pose loss vector during face recognition and the pose fusion feature in the facial contour recognition space.

[0038] In specific implementation, the matlab discrete drawing method in the prior art can be used to construct all the eigenvalues in the pose fusion feature into a grayscale image with 400*400 pixels of a scatter point, and use this grayscale image as the facial pose fusion map of the person. The facial pose fusion map represents the image obtained by mapping all the features in the pose fusion feature, and the facial pose fusion map can reflect all the initial features of the person's facial contour. Among them, deconvolving the pose fusion feature to obtain the local contour feature and mapping it into the facial contour recognition space is more conducive to subsequent processing and reduces the recognition error caused by the multi-form changes of the person at the same time.

[0039] In addition, in specific implementation, obtain the initial position of each facial key part combination map during face recognition, calculate the displacement and rotation angle between each facial key part combination map and the facial pose fusion map respectively, and linearly combine the calculated displacement and rotation angle to obtain the pose loss vector when the pose of the facial pose fusion map changes in each adjacent frame, that is, pose loss vector = (displacement, rotation angle), where the displacement can be determined by calculating the mean value of the slopes generated by the corresponding pixel points in all facial key part combination maps and the facial pose fusion map due to pose changes; then, randomly arrange all the pose loss vectors in a column to form a pose loss matrix, and then use the array composed of the mean value of the first column data and the mean value of the second column data of the pose loss matrix as the pose compensation array. It should be noted that the pose compensation array in this application refers to the standard array of the pose loss vectors of the facial key part combination maps, and the pose compensation array can represent the overall trend of the pose changes of the person's facial key parts.

[0040] In specific implementation, first, use the first value (i.e., the mean value of the displacements of all facial key part combination maps) and the second value (i.e., the mean value of the rotation angles of all facial key part combination maps) of the pose compensation array as the direction references for transposed convolution respectively; finally, perform transposed convolution on all the pose loss vectors through the set convolution kernel and direction references to obtain multiple local contour features of the person; then, use the Euclidean distance between each local contour feature in the facial contour recognition space and the facial pose fusion map as the displacement deviation of the corresponding local contour feature. The displacement deviation represents the feature difference of the target face caused by pose, expression and other factors when the pose fusion feature is mapped into the recognition space, and use the mean value of all the displacement deviations as the pose mapping cost during contour mapping.

[0041] It should be noted that the pose mapping cost in this application represents the deviation degree when the pose fusion feature is mapped to the local contour feature. The larger the pose mapping cost, the greater the deviation degree when the pose fusion feature is mapped to the local contour feature, and the smaller the pose mapping cost, the smaller the deviation degree when the pose fusion feature is mapped to the local contour feature. By correcting the changes in facial features caused by different postures, expressions, and other factors through the pose mapping cost, the error of face recognition can be reduced.

[0042] In step S4, the environmental light intensity during face recognition is obtained, and based on the environmental light intensity and the mapping cost, recognition compensation is performed on the face recognition process to obtain a compensated feature quantity.

[0043] Specifically, when implemented, the environmental light intensity during face recognition is obtained through the light sensor of the depth camera, that is, the light intensity sensor working together with the network camera is connected to obtain the environmental light intensity data corresponding to the face recognition period, and the average value of the environmental light intensity data during this period is used as the environmental light intensity when the network camera takes pictures.

[0044] In this embodiment, performing recognition compensation on the face recognition process based on the environmental light intensity and the mapping cost to obtain a compensated feature quantity specifically includes: Determining the feature correction quantity in the face recognition process according to the mapping cost; Determining the light compensation coefficient through the environmental light intensity; Obtaining the brightness reference value of the facial contour recognition space; Based on the feature correction quantity, the light compensation coefficient, and the brightness reference value, respectively perform feature adjustment and light compensation on the face recognition process, and then obtain a compensated feature quantity.

[0045] Specifically, when implemented, first, the feature correction quantity can be determined by the following formula, that is: feature correction quantity = 1 / (1 + mapping cost); then, read the overall brightness during face recognition from the video playback software, and use the ratio of the environmental light intensity to the overall brightness as the brightness error rate during face video acquisition. The brightness error rate represents the deviation degree between the environmental light intensity and the actual video brightness when acquiring the face video. The value range of the brightness error rate is between 0 and 1, that is, the larger the brightness error rate, the greater the deviation between the environmental light intensity and the actual video brightness, and the smaller the brightness error rate, the smaller the deviation between the environmental light intensity and the actual video brightness. Then, the light compensation coefficient can be determined by the difference between the preset facial occlusion standard and the brightness error rate. As a preferred embodiment, the value of the preset facial occlusion standard can be set to 1, that is: light compensation coefficient = 1 - brightness error rate.

[0046] It should be noted that the light compensation coefficient in this application represents the deviation between the ambient light intensity and the overall brightness of the face video. The light compensation coefficient can take into account the ambient light intensity when collecting the face video. By differentiating the ambient light intensity from the overall brightness of the face video, it can be determined whether there is occlusion in the collected face video, providing a basis for subsequent light compensation for facial contour features. Among them, the overall brightness of the face video can be the overall brightness of consecutive image frames.

[0047] In specific implementation, first, the adjust_brightness function in Python can be used to calculate the brightness reference value in the facial contour recognition space. The brightness reference value refers to the brightness standard in the facial contour space, and its value generally ranges from 100 to 180 nits. Then, light compensation is performed on the face recognition process through the light compensation coefficient and the brightness reference value respectively. Among them, the light compensation amount can be adjusted based on the ambient light intensity and the light compensation coefficient to the brightness reference value. As a preferred embodiment, the light compensation amount can be determined by the following formula, that is: light compensation amount = brightness reference value - ambient light intensity × light compensation coefficient. Among them, performing light compensation can improve the recognition accuracy of the face recognition system under different lighting conditions. Finally, the light compensation amount and the feature correction amount are weighted and summed to obtain the compensated feature amount, which can improve the accuracy of face recognition.

[0048] It should be noted that the compensated feature amount in this application refers to the fusion amount of feature adjustment and light compensation. Among them, by adjusting the brightness of the face recognition process through the light compensation amount, the influence of light changes on facial features can be reduced, enabling the face recognition to still extract image features with consistent brightness even under large changes in light intensity, thereby improving the accuracy of facial recognition. Through feature adjustment, the mapping cost in the face recognition process can be reduced, which is beneficial to improving the accuracy rate of face recognition.

[0049] In step S5, the matching confidence of face recognition is determined through the compensated feature amount, and then the face recognition results are screened based on the matching confidence.

[0050] In this embodiment, determining the matching confidence of face recognition through the compensated feature amount and then screening the face recognition results based on the matching confidence can be specifically implemented in the following way, that is: Obtain all the face features to be matched in the database of the face recognition system; Determine the matching confidence of face recognition through all the face features to be matched and the compensated feature amount; Match and screen the face recognition results according to the matching confidence of face recognition.

[0051] In specific implementation, first, connect to the database of the face recognition system and use the existing feature extraction model to extract the face features to be matched of all the face images to be matched in the database. Here, the database can be a public database or can be established according to the actual application scenario, which is not limited here. The database includes multiple preset face images to be matched. Then, an existing face recognition model (such as: VGGFace, FaceNet, DeepFace, ArcFace, OpenFace, etc.) can be used to use the compensation feature amount as a pre-adjustment in the face recognition process, and then match the adjusted image features with the face features to be matched of each face image to be matched in the database, and calculate the matching confidence of each face feature to be matched through the corresponding confidence function in the face recognition model, so as to use the face image corresponding to the highest matching confidence as the face recognition result of the person.

[0052] It should be noted that when all the matching confidences are less than 75%, mark the face image corresponding to the highest matching confidence as the suspected matching result of the face and prompt that further verification is required, indicating that there is no face image to be matched in the database. The face images to be matched can be added according to the actual application scenario or the database selected during face recognition can be replaced.

[0053] It can be seen that in this application, dynamic feature adjustment and light compensation in the face recognition process can be realized. First, by acquiring continuous image frames, the dynamic features of the face can be captured, reducing the recognition error caused by instantaneous changes, providing more image feature information, and reflecting the change trend of the face in a short time through the inter-frame difference vector, facilitating dynamic correction of the pose error to reduce the feature deviation caused by head rotation or tilt, making the feature extraction more accurate and improving the face matching accuracy under different poses. Secondly, by adjusting the pose fusion features through the spatial difference angle to map them to the standardized facial contour recognition space, the influence caused by individual differences or shooting angle changes can be reduced. Using the deconvolution operation can restore the local contour details, making the facial features clearer and conducive to improving the accuracy of feature matching. Then, by acquiring the ambient light information in real time, the images in low-light or strong-light conditions can be dynamically adjusted, and compensation can be carried out in combination with the pose mapping cost, which can effectively reduce the influence of light changes on feature extraction and improve the adaptability of the system to different light conditions. Finally, the final matching result of face recognition can be optimized through the compensation feature amount, improving the adaptability of the system to dynamic changes, which is conducive to reducing the recognition error and improving the overall accuracy of face recognition.

[0054] In summary, the technical solution adopted in this application can realize dynamic feature adjustment and light compensation in the face recognition process to improve the accuracy of face recognition.

[0055] Embodiment 2. The present application provides a face recognition system based on image analysis. Refer to Figure 4 As shown in the figure, which is a schematic diagram of the face recognition system according to this embodiment of the present application. The face recognition system includes: An image acquisition module 100, configured to acquire consecutive image frames during face recognition; A pose fusion module 200, configured to determine an inter-frame difference vector according to the difference information between adjacent frames in the consecutive image frames, and perform pose fusion on the consecutive image frames through all the inter-frame difference vectors to obtain a pose fusion feature; A feature processing module 300, configured to map the pose fusion feature to a facial contour recognition space based on the spatial difference angle during face recognition, perform deconvolution on the pose fusion feature in the facial contour recognition space, and further determine a pose mapping cost during face recognition from the local contour features obtained by the deconvolution; A light compensation module 400, configured to obtain the environmental light intensity during face recognition, and perform recognition compensation on the face recognition process according to the environmental light intensity and the mapping cost to obtain a compensated feature quantity; A matching and screening module 500, configured to determine the matching confidence of face recognition through the compensated feature quantity, and further screen the face recognition result according to the matching confidence.

[0056] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowcharts and / or block diagrams can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0057] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0058] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

Claims

1. A face recognition method based on image analysis, characterized in that: The face recognition method comprises: Collect continuous image frames during face recognition; Determine an inter-frame difference vector according to differential information between adjacent frames in the continuous image frames, perform posture fusion on the continuous image frames through all the inter-frame difference vectors, and obtain a posture fusion feature; Mapping the posture fusion feature to the facial contour recognition space based on the spatial difference angle during face recognition, deconvolving the posture fusion feature in the facial contour recognition space, and then determining the posture mapping cost during face recognition based on the local contour feature obtained by deconvolution; Acquire the ambient light intensity during face recognition, perform recognition compensation on the face recognition process according to the ambient light intensity and the mapping cost, and obtain a compensation feature value; The matching confidence of face recognition is determined by the compensation feature quantity, and then the face recognition result is screened according to the matching confidence.

2. A face recognition method based on image analysis as claimed in claim 1, characterized in that: The depth camera is used to collect continuous image frames for face recognition.

3. A face recognition method based on image analysis as claimed in claim 1, characterized in that: Determining the inter-frame difference vector according to the difference information between adjacent frames in the continuous image frames specifically includes: Extracting a key feature map in each image frame from the continuous image frames; Determine the difference images of adjacent frames in the continuous image frames, and then determine the inter-frame difference vectors of the adjacent frames based on the difference images of the adjacent frames and all key feature maps.

4. A face recognition method based on image analysis as claimed in claim 1, characterized in that: The continuous image frames are subjected to posture fusion through all inter-frame difference vectors, and the posture fusion features obtained specifically include: Construct the posture fusion matrix based on all inter-frame difference vectors; The gesture fusion feature is determined by the eigenvalue of the gesture fusion matrix.

5. The face recognition method based on image analysis as claimed in claim 1, characterized in that: Mapping the posture fusion feature to the facial contour recognition space based on the spatial difference angle during face recognition specifically includes: Determine a mapping angle according to the spatial difference angle; The posture fusion feature is linearly mapped based on the mapping angle to obtain a facial contour recognition space.

6. A face recognition method based on image analysis as claimed in claim 1, characterized in that: The posture mapping cost for face recognition determined by the local contour features obtained by deconvolution specifically includes: Determine a facial posture fusion graph according to the posture fusion features in the facial contour recognition space; Determine the posture loss vector during face recognition by using the facial posture fusion graph and all local contour features; The posture mapping cost during face recognition is determined by performing deconvolution on the posture loss vector during face recognition and the posture fusion feature in the facial contour recognition space.

7. A face recognition method based on image analysis as claimed in claim 1, characterized in that: Deconvolution is performed on the posture fusion features in the facial contour recognition space through a preset neural network model.

8. The face recognition method based on image analysis as claimed in claim 1, characterized in that: The depth camera's illumination sensor is used to obtain the ambient light intensity during face recognition.

9. The face recognition method based on image analysis as claimed in claim 1, characterized in that: The face recognition process is recognized and compensated according to the ambient light intensity and the mapping cost, and the compensation feature quantity is obtained, which specifically includes: Determining a feature correction amount in a face recognition process according to the mapping cost; Determine the illumination compensation coefficient according to the ambient illumination intensity; Obtaining a brightness reference value of a facial contour recognition space; Based on the feature correction amount, the illumination compensation coefficient and the brightness reference value, feature adjustment and light compensation are performed on the face recognition process respectively, so as to obtain the compensated feature amount.

10. A face recognition system based on image analysis, used to execute a face recognition method based on image analysis as claimed in any one of claims 1 to 9, characterized in that: The face recognition system comprises: An image acquisition module, used to acquire continuous image frames during face recognition; A posture fusion module, used to determine an inter-frame difference vector according to differential information between adjacent frames in the continuous image frames, and perform posture fusion on the continuous image frames through all the inter-frame difference vectors to obtain a posture fusion feature; A feature processing module, used to map the posture fusion feature to a facial contour recognition space based on a spatial difference angle during face recognition, deconvolute the posture fusion feature in the facial contour recognition space, and then determine the posture mapping cost during face recognition from the local contour feature obtained by deconvolution; An illumination compensation module is used to obtain the ambient illumination intensity during face recognition, and to perform recognition compensation on the face recognition process according to the ambient illumination intensity and the mapping cost to obtain a compensation feature value; The matching and screening module is used to determine the matching confidence of face recognition through the compensation feature quantity, and then screen the face recognition results according to the matching confidence.

Images