Risk control identity authentication system and method based on facial expression recognition

Abstract

The invention discloses a risk control identity authentication system and method based on facial expression recognition, and relates to the technical field of image recognition. The system comprises an identity information acquisition module, a facial feature extraction module, a dynamic feature extraction accuracy analysis module, a multi-dimensional comparison module and a multi-dimensional comparison rationality analysis module. Dynamic feature extraction accuracy evaluation is obtained through feature extraction parameter quantification, whether dynamic feature extraction accuracy optimization is carried out or not is judged, and if yes, a facial expression feature comparison link is carried out after optimization; otherwise, a facial expression feature comparison link is directly carried out, multi-dimensional comparison parameter quantification is obtained to obtain multi-dimensional comparison rationality evaluation, and whether multi-dimensional comparison rationality optimization is carried out is judged, so that the risk control identity authentication reliability is improved, and the problem that in the prior art, the risk control identity authentication efficiency is high is solved. The problem of low risk control identity authentication reliability caused by inaccurate expression dynamic feature extraction exists.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a risk control identity authentication system and method based on facial expression recognition. Background Art

[0002] Users follow prompts (such as the screen prompting "Smile and blink") to make designated expressions. The acquisition device captures the dynamic expression video stream or sequence image in real time. At the same time, infrared imaging and 3D depth detection are used to determine whether the user is a real person to prevent forgery attacks such as photos, videos, and masks. The collected dynamic data is preprocessed (such as denoising, face alignment, and keyframe extraction). Computer vision algorithms such as deep learning-based CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory) models are used to extract dynamic features of expressions such as facial muscle motion vectors, expression duration, and temporal changes in feature points, as well as static features such as facial geometric structure.

[0003] The facial expression features extracted in real time are compared with the facial expression-identity benchmark model stored when the user registered in multiple dimensions to calculate the matching degree (such as the temporal consistency of dynamic features, the similarity of key facial expression movements, the overlap of feature point motion trajectories, etc.). If all verification links are passed, the system will feedback to the user that the identity authentication is successful and allow them to continue the operation; if it fails, the reason for the failure will be clearly prompted, such as insufficient expression matching, detection of non-liveness, and the user will be guided to complete the identity authentication through other secure methods (such as SMS verification code, fingerprint recognition).

[0004] For example, the identity authentication method, device, equipment and system disclosed in the Chinese invention patent with announcement number: CN105809111B include: obtaining the user's ID information; taking the user's facial photo; judging whether the camera is facing a real person based on the user's head image obtained in real time by the camera; and when it is determined that the camera is facing a real person, sending the user's ID information and facial photo in association to the authentication server.

[0005] For example, a Chinese invention patent with publication number CN108875646A discloses a dual comparison authentication method and system for a real facial image and an ID card registration, comprising: obtaining a current facial image and a registered facial image of a target person; performing feature extraction on the current facial image and the registered facial image respectively to generate corresponding texture feature vectors; correcting the texture feature vector of the registered facial image using a pre-set correction model to generate a corrected feature vector; calculating the difference between the texture feature vector of the current facial image and the corrected feature vector, and determining whether the difference value is less than a preset threshold; and determining that the current facial image and the registered facial image match when the difference value is less than the preset threshold.

[0006] The above technology has at least the following technical problems: During the risk control identity authentication process, dynamic facial features are extracted through CNN and LSTM. For the LSTM model, its core is to capture the temporal changes of facial expressions, which relies on the integrity of continuous key frames. Blurred frames and lost frames caused by sudden changes in ambient light or network packet loss will destroy the continuity of facial movements, making it impossible for the LSTM model to accurately extract the temporal changes of facial expressions, which in turn leads to deviations in dynamic feature extraction. The CNN model is easily affected by changes in makeup appearance. The differences in user registration and recognition, such as bare face / makeup, beard / glasses, can reduce the facial feature matching score. At the same time, it is sensitive to lighting and posture. Side light causes distortion in the calculation of key features, affecting the CNN model's accurate extraction of static features and further exacerbating the inaccuracy of dynamic feature extraction.

[0007] Multi-dimensional comparison has significant accuracy defects. On the one hand, there is a logical imbalance in the weight distribution mechanism. Some systems over-rely on a single indicator (such as setting a high weight for "expression switching speed"), but fail to consider the natural movement differences of users due to tension, fatigue or emotional fluctuations. On the other hand, the deformation of the chin-neck trajectory caused by the forward tilt of the head when sitting and the physiological differences in pupil reflection intensity under different lighting are all predictable scene variables, but the existing system has not designed a dynamic threshold adjustment mechanism for these variables, resulting in systematic deviations in the feature comparison results of the same user in different scenarios. There is a problem of low reliability of risk control identity authentication due to inaccurate extraction of dynamic expression features. Summary of the Invention

[0008] To address the problem of low reliability of risk control identity authentication due to inaccurate facial expression dynamic feature extraction in the prior art, the present invention provides a risk control identity authentication system and method based on facial expression recognition. The technical solution is as follows: On the one hand, a risk control identity authentication system based on facial expression recognition is provided, comprising the following steps: an identity information acquisition module, a facial feature extraction module, a dynamic feature extraction accuracy analysis module, a multi-dimensional comparison module and a multi-dimensional comparison rationality analysis module: wherein the identity information acquisition module is used to acquire a facial image sequence of a user to be authenticated; the facial feature extraction module is used to pre-process the facial image sequence to obtain feature extraction parameters to quantify the accuracy of CNN and LSTM models in the facial expression feature extraction process, and obtain a dynamic feature extraction accuracy evaluation; the dynamic feature extraction accuracy analysis module is used to determine whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy evaluation, and if so, to determine whether to perform dynamic feature extraction accuracy optimization during the optimization process. Then the facial expression feature comparison link is performed, otherwise the facial expression feature comparison link is performed directly. The dynamic feature extraction accuracy optimization includes dynamic adjustment of regional feature weights and dynamic adjustment of facial area mask weights; the multi-dimensional comparison module is used to obtain the multi-dimensional comparison parameters in the facial expression feature comparison link to quantify the rationality of the facial expression feature comparison process and obtain a multi-dimensional comparison rationality evaluation; the multi-dimensional comparison rationality analysis module is used to determine whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality evaluation. If so, the risk control identity authentication decision is generated after optimization, otherwise the risk control identity authentication decision is generated directly. The multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights and dynamic adjustment of static-dynamic feature weights.

[0009] On the other hand, a risk control identity authentication method based on facial expression recognition is provided, including: collecting a facial image sequence of a user to be authenticated; preprocessing the facial image sequence to obtain feature extraction parameters to quantify the accuracy of CNN and LSTM models in the facial expression feature extraction process, and obtaining a dynamic feature extraction accuracy evaluation; judging whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy evaluation, and if so, performing a facial expression feature comparison link after optimization, otherwise directly performing a facial expression feature comparison link, and the dynamic feature extraction accuracy optimization includes dynamically adjusting the regional feature weights to improve the discrimination in facial expression feature extraction, And the facial area mask weight is dynamically adjusted to improve the accuracy of feature representation; the multi-dimensional comparison parameters in the facial expression feature comparison link are obtained to quantify the rationality of the facial expression feature comparison process, and obtain a multi-dimensional comparison rationality evaluation; based on the multi-dimensional comparison rationality evaluation, it is determined whether to perform multi-dimensional comparison rationality optimization. If so, the risk control identity authentication decision is generated after optimization, otherwise the risk control identity authentication decision is directly generated. The multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights to improve the matching accuracy of dynamic feature sequences during expression switching, and dynamic adjustment of static-dynamic feature weights to improve the consistency between multi-dimensional comparison and real identity.

[0010] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By collecting facial image sequences of the user to be authenticated and preprocessing them to obtain feature extraction parameters, clear and standardized facial image data can be obtained, providing a high-quality input basis for subsequent facial feature extraction and identity authentication, and effectively reducing the impact of interference factors on authentication accuracy. Based on the quantification of feature extraction parameters, a dynamic feature extraction accuracy evaluation is obtained to characterize the accuracy of CNN and LSTM models in the facial expression feature extraction process. Based on the dynamic feature extraction accuracy evaluation, it is determined whether to perform dynamic feature extraction accuracy optimization to achieve dynamic control and targeted improvement of accuracy in the facial expression feature extraction process, thereby improving the reliability of facial feature extraction. The multi-dimensional comparison parameters in the facial expression feature comparison link are quantified to obtain a multi-dimensional comparison rationality evaluation to characterize the rationality of the facial expression feature comparison process. Based on the multi-dimensional comparison rationality evaluation, it is determined whether to perform multi-dimensional comparison rationality optimization to achieve dynamic control and targeted optimization of the entire multi-dimensional comparison process, thereby comprehensively improving the reliability of risk control identity authentication.

[0011] 2. Based on the dynamic feature extraction accuracy assessment, determine whether to optimize dynamic feature extraction accuracy. If so, determine whether to dynamically adjust regional feature weights based on brightness deviation. Specifically, brightness deviation reflects the uniformity of illumination across different regions of the facial image. When certain regions are too bright or too dark, the expression features in these regions may be masked or misinterpreted, affecting the accuracy of feature extraction. By incorporating brightness deviation into the decision to adjust regional feature weights, the model can assign higher weights to regions with suitable illumination and clear features during the feature extraction process, giving them a more prominent position in feature calculation. It also appropriately reduces the weights of regions with blurred features due to brightness deviation, reducing their interference with the overall feature extraction results. This significantly improves the accuracy and discrimination of facial expression feature extraction. Furthermore, based on mask entropy, determine whether to dynamically adjust facial region mask weights. This can specifically optimize the mask's ability to focus on valid feature regions. By increasing the mask weights of valid feature regions and reducing the weights of irrelevant regions, the model can more accurately capture key expression features, reduce redundant information interference, and improve the accuracy of feature representation, thereby comprehensively improving the reliability of risk control identity authentication.

[0012] 3. Determine whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality assessment. If so, determine whether to dynamically adjust the transition frame weights based on the expression switching interval. This avoids the interference of invalid transition information and makes full use of effective transition features, thereby significantly improving the rationality, accuracy and robustness of the multi-dimensional comparison, providing more reliable support for the precise matching of facial expression features, and determining whether to dynamically adjust the static-dynamic feature weights based on the positioning error of a single feature point. This avoids the interference of invalid transition information, makes full use of effective transition features, and improves the rationality and accuracy of the comparison. Further, determine whether to dynamically adjust the static-dynamic feature weights based on the positioning error of a single feature point, thereby comprehensively improving the reliability of risk control identity authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A schematic diagram of the structure of a risk control identity authentication system based on facial expression recognition provided in an embodiment of the present application; Figure 2 A flowchart of dynamic adjustment of facial area mask weights in a risk control identity authentication system based on facial expression recognition provided in an embodiment of the present application; Figure 3 A flowchart of the dynamic adjustment of transition frame weights in a risk control identity authentication system based on facial expression recognition provided in an embodiment of the present application; Figure 4 A flow chart of a risk control identity authentication method based on facial expression recognition provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0016] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0017] The embodiment of the present application solves the problem of low reliability of risk control identity authentication in the prior art due to inaccurate extraction of dynamic features of facial expressions by providing a risk control identity authentication system and method based on facial expression recognition. By collecting a facial image sequence of the user to be authenticated and preprocessing it to obtain feature extraction parameters, the accuracy evaluation of dynamic feature extraction is obtained, and it is determined whether to optimize the accuracy of dynamic feature extraction. If so, a facial expression feature comparison link is performed after optimization. Otherwise, the facial expression feature comparison link is directly performed and the multi-dimensional comparison parameters are quantified to obtain a multi-dimensional comparison rationality evaluation, and it is determined whether to perform multi-dimensional comparison rationality optimization, thereby improving the reliability of risk control identity authentication.

[0018] The technical solution in the embodiments of the present application is to solve the problem of low reliability of risk control identity authentication caused by inaccurate extraction of facial expression dynamic features. The overall idea is as follows: By collecting the facial image sequence of the user to be authenticated, the facial image sequence is preprocessed to obtain feature extraction parameters to quantify the accuracy of CNN and LSTM models in the facial expression feature extraction process, and obtain a dynamic feature extraction accuracy evaluation. According to the dynamic feature extraction accuracy evaluation, it is judged whether to perform dynamic feature extraction accuracy optimization. If so, the facial expression feature comparison link is performed after optimization, otherwise the facial expression feature comparison link is directly performed, and the multi-dimensional comparison parameters in the facial expression feature comparison link are obtained to quantify the rationality of the facial expression feature comparison process, and obtain a multi-dimensional comparison rationality evaluation. According to the multi-dimensional comparison rationality evaluation, it is judged whether to perform multi-dimensional comparison rationality optimization. If so, the risk control identity authentication decision is generated after optimization, otherwise the risk control identity authentication decision is directly generated, thereby improving the reliability of risk control identity authentication.

[0019] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1As shown, it is a structural diagram of a risk control identity authentication system based on facial expression recognition provided by an embodiment of the present application. The processing flow of the system may include the following steps: an identity information acquisition module, a facial feature extraction module, a dynamic feature extraction accuracy analysis module, a multi-dimensional comparison module and a multi-dimensional comparison rationality analysis module, and the identity information acquisition module, whose output end is connected to the input end of the facial feature extraction module; the facial feature extraction module, whose first output end is connected to the first input end of the multi-dimensional comparison module, and whose second output end is connected to the input end of the dynamic feature extraction accuracy analysis module; the dynamic feature extraction accuracy analysis module, whose output end is connected to the second input end of the multi-dimensional comparison module; the multi-dimensional comparison module, whose third input end is connected to the identity authentication database, and whose output end is connected to the input end of the multi-dimensional comparison rationality analysis module; the multi-dimensional comparison rationality analysis module, whose data end is connected to the identity authentication database.

[0021] Among them, the identity information acquisition module is used to collect the facial image sequence of the user to be authenticated; the facial feature extraction module is used to preprocess the facial image sequence to obtain feature extraction parameters to quantify the accuracy of the CNN and LSTM models in the facial expression feature extraction process, and obtain a dynamic feature extraction accuracy evaluation.

[0022] It should be noted that the data of this application will be normalized and only its numerical results will be considered. The feature extraction parameters include: key frame missing rate, single feature point displacement vector field and frame interval time; the feature extraction parameter threshold and feature extraction parameter correction factor are obtained from the constructed identity authentication database. The feature extraction parameter threshold includes the key frame missing rate threshold, the single feature point displacement vector field threshold and the frame interval time standard value. The feature extraction parameter correction factor includes the key frame missing rate correction factor, the single feature point displacement vector field correction factor and the frame interval time correction factor. Specifically, the number of key frames is extracted through video parsing tools such as FFmpeg, and the ratio of the absolute value of the difference between the number of key frames and the preset number of key frames to the preset number of key frames is recorded as the key frame missing rate. A key frame refers to a special frame containing a frame of expression image information; key feature points are located in continuous video frames through facial detection algorithms such as MTCNN, the coordinates of the feature points in each frame are recorded, the direction of the feature points is calculated by inverse tangent, and then the coordinate difference of the same feature point in two adjacent frames is calculated by the Euclidean distance formula to obtain the displacement vector of a single feature point, which is recorded as a single feature point displacement vector field; the time interval between two adjacent frames of images is obtained through the hardware clock synchronization of the acquisition device, which is recorded as the frame interval time.

[0023] Regarding the above content, it should be noted that in order to support the efficient operation of a risk control identity authentication system based on facial expression recognition, the system has pre-built an identity authentication database specifically for storing various key setting parameters, including key frame loss rate, single feature point displacement vector field, frame interval time and other system-required preset data. The initial value setting of these parameters is not subjective, but based on actual data support - taking the key frame loss rate threshold as an example, its initial value is calculated by summing and averaging the historical key frame loss rates accumulated in the database, ensuring that the initial setting has an objective basis and representativeness. At the same time, considering the complexity of actual application scenarios and the dynamic changes in demand, all parameter values ​​in the database remain adjustable. Technical personnel can manually set, adjust and fine-tune these values ​​according to the performance of the system during the actual debugging process, and ultimately achieve the optimal configuration of system parameters.

[0024] It should be explained that, first, the key frame missing rate threshold and the key frame missing rate ratio analysis result are corrected by the key frame missing rate correction factor, which is recorded as the first component of accuracy. Among them, the ratio analysis represents a division operation, and the specific restriction expression is: ; Where Q represents the first accuracy component, P1 represents the keyframe loss rate correction factor obtained from the authentication database, A0 represents the keyframe loss rate threshold obtained from the authentication database, A represents the keyframe loss rate, and k1 represents a constant term to avoid meaningless numbers. The keyframe loss rate is negatively correlated with the dynamic feature extraction accuracy assessment. A higher keyframe loss rate indicates more keyframes are lost during video transmission, storage, or processing, and the dynamic feature extraction accuracy assessment decreases.

[0025] Next, the analysis result of the ratio of the single feature point displacement vector field to the single feature point displacement vector field threshold is corrected by the single feature point displacement vector field correction factor, which is recorded as the second component of accuracy. The specific restriction expression is: ; Where L represents the second component of accuracy, P2 represents the correction factor for the displacement vector field of a single feature point obtained from the identity verification database, B represents the displacement vector field of a single feature point, and B0 represents the threshold of the displacement vector field of a single feature point obtained from the identity verification database. The displacement vector field of a single feature point is positively correlated with the accuracy of dynamic feature extraction. A higher displacement vector field value indicates that the movement trajectory of various facial parts can be accurately described, providing rich dynamic features such as velocity, acceleration, and direction changes, and the dynamic feature extraction accuracy is higher.

[0026] Then, the analysis result of the ratio of the standard value of the frame interval time and the interval time deviation is corrected by the frame interval time correction factor, which is recorded as the third component of accuracy. Among them, the interval time deviation represents the absolute value of the difference between the frame interval time and the standard value of the frame interval time. The specific restriction expression is: ; Where D represents the third accuracy component, P3 represents the frame interval correction factor obtained from the authentication database, R0 represents the standard frame interval value obtained from the authentication database, and R represents the frame interval. There is a negative correlation between the frame interval and the accuracy of dynamic feature extraction. A longer frame interval means a lower sampling rate, which weakens the ability to capture dynamic changes in expression. This results in the loss of intermediate transition frames, an inability to accurately describe the trajectory of expression changes, and a lower dynamic feature extraction accuracy assessment.

[0027] Finally, the first accuracy component, the second accuracy component, and the third accuracy component are coupled to obtain the accuracy evaluation of dynamic feature extraction. The specific restriction expression is: ; Where U represents the accuracy evaluation of dynamic feature extraction.

[0028] It's important to explain that the authentication database stores influencing factors related to feature extraction parameters, including the keyframe loss rate correction factor, the single feature point displacement vector field correction factor, and the frame interval correction factor. The values ​​of these influencing factors mostly range from 0 to 1, and the sum of the three values ​​is 1. There is also a pre-defined mapping between these influencing factors and feature extraction parameters. This relationship can be a one-to-one direct correspondence or a many-to-one aggregated correspondence.

[0029] Specifically, there are correlations between feature extraction parameters, as follows: the longer the frame interval, the smaller the speed at the same displacement distance, and the smaller the motion speed, and the smaller the displacement vector field of a single feature point; the higher the key frame missing rate, the failure of feature point matching, the absence of a reference frame, and the smaller the displacement vector field of a single feature point.

[0030] It should be further explained that understanding the correlation between feature extraction parameters helps to predict the impact of different parameter changes on the displacement vector field of feature points in advance during the actual extraction process: when the frame interval is detected to increase, the sensitivity of the feature tracking algorithm can be adjusted in a targeted manner to avoid misjudgment of the displacement vector field due to speed calculation deviation. Adaptive adjustments can be made based on its correlation with other parameters to ensure the consistency and reliability of the feature extraction results, providing better basic data for subsequent facial expression feature comparison and identity authentication. Understanding the positive and negative correlation between feature extraction parameters and dynamic feature extraction accuracy assessment helps to achieve fine-grained control and optimization of the feature extraction process in actual operation, accurately identify key parameters that affect accuracy and their direction of action, avoid waste of resources or fluctuations in results caused by blind adjustments, and thus achieve a continuous and stable improvement in the accuracy of dynamic feature extraction, output more reliable feature data for subsequent facial expression feature comparison and identity authentication, and fundamentally enhance the performance of the entire authentication system.

[0031] The dynamic feature extraction accuracy analysis module is used to determine whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy evaluation. If so, the facial expression feature comparison link is performed after optimization; otherwise, the facial expression feature comparison link is performed directly. The dynamic feature extraction accuracy optimization includes dynamic adjustment of regional feature weights to improve the discrimination in facial expression feature extraction, and dynamic adjustment of facial region mask weights to improve the accuracy of feature representation.

[0032] The multi-dimensional comparison module is used to obtain the multi-dimensional comparison parameters in the facial expression feature comparison link to quantify the rationality of the facial expression feature comparison process and obtain a multi-dimensional comparison rationality evaluation.

[0033] It should be noted that the multi-dimensional comparison parameters include: the dynamic feature extraction accuracy assessment to be compared, the single feature point positioning error, and the expression switching interval. Specifically, if the dynamic feature extraction accuracy is optimized, the optimized dynamic feature extraction accuracy assessment is recorded as the dynamic feature extraction accuracy assessment to be compared; otherwise, the current dynamic feature extraction accuracy assessment is directly obtained and recorded as the dynamic feature extraction accuracy assessment to be compared. A feature point detection algorithm such as SIFT is used to process the image, extracting and outputting the coordinates of the located feature points. The absolute value of the difference between the located feature point coordinates and the preset coordinates calculated using Euclidean distance is recorded as the single feature point positioning error. The time interval required for the user to switch from one expression to another is recorded using a high-speed camera as the expression switching interval. Multi-dimensional comparison parameter thresholds and multi-dimensional comparison parameter correction factors are obtained from the constructed authentication database. The multi-dimensional comparison parameter thresholds include the dynamic feature extraction accuracy assessment threshold, the single feature point positioning error standard value, and the expression switching interval standard value. The multi-dimensional comparison parameter correction factors include the dynamic feature extraction accuracy assessment correction factor, the single feature point positioning error correction factor, and the expression switching interval correction factor.

[0034] It should be explained that, first, the analysis results of the proportion of the dynamic feature extraction accuracy evaluation to be compared and the dynamic feature extraction accuracy evaluation threshold are corrected by the dynamic feature extraction accuracy evaluation correction factor, which is recorded as the first rationality component. The specific restriction expression is: ; Where V represents the first rationality component, C1 represents the correction factor for the dynamic feature extraction accuracy assessment obtained from the identity verification database, F0 represents the dynamic feature extraction accuracy assessment threshold obtained from the identity verification database, and F represents the dynamic feature extraction accuracy assessment for comparison. There is a positive correlation between the dynamic feature extraction accuracy assessment for comparison and the rationality assessment of the multi-dimensional comparison. A higher dynamic feature extraction accuracy assessment for comparison indicates that temporal features, such as AU intensity changes and expression transition trajectories, are closer to reality, and the rationality assessment of the multi-dimensional comparison is higher.

[0035] Next, the analysis results of the proportion of the standard value of the positioning error of a single feature point and the degree of positioning error deviation are corrected by the single feature point positioning error correction factor. The degree of positioning error deviation represents the absolute value of the difference between the positioning error of a single feature point and the standard value of the positioning error of a single feature point, which is recorded as the second component of rationality. The specific restriction expression is: ; Where S represents the second rationality component, C2 represents the correction factor for the positioning error of a single feature point obtained from the authentication database, W0 represents the standard value of the positioning error of a single feature point obtained from the authentication database, and W represents the positioning error of a single feature point. The positioning error of a single feature point is negatively correlated with the rationality assessment of the multi-dimensional comparison. Higher positioning errors lead to deviations in the spatial dimension comparison and lower the rationality assessment of the multi-dimensional comparison.

[0036] Then, the analysis results of the ratio of the standard value of the expression switching interval to the deviation degree of the switching interval are corrected by the expression switching interval correction factor, which is recorded as the third component of rationality. Among them, the switching interval deviation degree represents the absolute value of the difference between the expression switching interval and the standard value of the expression switching interval. The specific restriction expression is: ; Where N represents the third component of rationality, C3 represents the expression switching interval correction factor obtained from the identity verification database, H0 represents the standard value of the expression switching interval obtained from the identity verification database, and H represents the expression switching interval. The expression switching interval is negatively correlated with the rationality assessment of multi-dimensional comparison. Longer expression switching intervals lead to loss of dynamic information and feature loss, resulting in a lower rationality assessment of multi-dimensional comparison.

[0037] It's important to explain that the authentication database stores influencing factors related to multi-dimensional comparison parameters. Specifically, these factors include a correction factor for dynamic feature extraction accuracy, a correction factor for single feature point positioning errors, and a correction factor for expression switching intervals. The values ​​of these influencing factors mostly range from 0 to 1, and the sum of the three values ​​is 1. Furthermore, there is a pre-defined mapping relationship between these influencing factors and multi-dimensional comparison parameters. This relationship can be either a one-to-one direct correspondence or a many-to-one aggregated correspondence.

[0038] Specifically, there are correlations between the multi-dimensional comparison parameters, as follows: the higher the dynamic feature extraction accuracy assessment, the higher the accuracy in the facial expression feature extraction process, and the smaller the positioning error of a single feature point; the shorter the expression switching interval, the shorter the time required to capture expression changes, and the higher the sampling density and real-time requirements for dynamic features, and the higher the dynamic feature extraction accuracy assessment; the larger the single feature point positioning error, the more it forces the system to extend the judgment threshold of the expression switching interval to avoid misjudgment, and the longer the expression switching interval.

[0039] It should be further explained that by understanding the correlation between multi-dimensional comparison parameters, it is helpful to reduce the contradictions and deviations caused by isolated parameter adjustments, and to improve the system's adaptability to different expression change scenarios, and provide a more reliable mechanism support for the accurate matching of facial expression features, thereby enhancing the stability and effectiveness of the entire identity authentication process; by understanding the positive and negative correlation between multi-dimensional comparison parameters and multi-dimensional comparison rationality assessment, it is helpful to quickly locate parameters that may cause fluctuations and make targeted adjustments, ensuring that multi-dimensional comparisons always maintain a high level of rationality, providing a more solid guarantee for the accurate matching of facial expression features, and thereby promoting the entire identity authentication process to develop in a more efficient and reliable direction.

[0040] The multi-dimensional comparison rationality analysis module is used to determine whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality assessment. If so, the risk control identity authentication decision is generated after optimization; otherwise, the risk control identity authentication decision is directly generated. The risk control identity authentication decision includes identity authentication passed and identity authentication failed. The multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights to improve the matching accuracy of dynamic feature sequences during expression switching, and dynamic adjustment of static-dynamic feature weights to improve the consistency of multi-dimensional comparison with the real identity.

[0041] As an embodiment of the first aspect, the reliability of risk control identity authentication is significantly improved through multi-module collaborative work and dynamic optimization mechanism. The closed-loop design of dynamic optimization and multi-dimensional evaluation effectively solves the problems of insufficient feature extraction discrimination, large dynamic sequence matching errors, and unreasonable fusion of static and dynamic features in traditional facial expression authentication, making the generated risk control identity authentication decisions (including identity authentication pass and fail) more accurate and reliable, and comprehensively improving the performance of the entire risk control identity authentication system.

[0042] Furthermore, the specific judgment steps for determining whether to perform dynamic feature extraction accuracy optimization are as follows: obtaining the accuracy assessment threshold from the constructed identity authentication database; if the dynamic feature extraction accuracy assessment is not less than the accuracy assessment threshold, then the dynamic feature extraction accuracy optimization is not performed, and a feature extraction accuracy qualification instruction is sent to the multi-dimensional comparison module; if the dynamic feature extraction accuracy assessment is less than the accuracy assessment threshold, then judging whether to perform dynamic adjustment of the regional feature weight according to the brightness deviation, and if so, judging whether to perform dynamic adjustment of the facial area mask weight after the dynamic adjustment, otherwise directly judging whether to perform dynamic adjustment of the facial area mask weight.

[0043] As a further explanation, the specific process for determining whether to dynamically adjust regional feature weights based on brightness deviation is as follows: The preset shadow brightness upper limit, preset shadow brightness lower limit, preset highlight brightness lower limit, and preset highlight brightness upper limit are obtained from the established identity authentication database. The preset shadow brightness upper limit is less than the preset shadow brightness lower limit, the preset shadow brightness lower limit is less than the preset highlight brightness lower limit, and the preset highlight brightness lower limit is less than the preset highlight brightness upper limit. The image is converted from the RGB (Red, Green, Blue) color space to the HSV (Hue, Saturation, Value) color space, and the V channel (Value) is extracted. The image is then divided into regions according to preset key areas (such as the corners of the mouth, wings of the nose, and jawline). The V values ​​of all pixels in each region are averaged to obtain the actual brightness value of the region. The difference between the actual brightness value and the preset reference brightness value is recorded as the brightness deviation. It should be explained that the highlights in non-critical areas such as the forehead and cheeks can be adjusted directly according to the threshold, while the highlights in the eyes (such as pupil reflections) need to be processed separately in combination with pupil detection (to avoid being misjudged as highlight areas), and the adjusted regional weights are converted into a "pixel-level weight matrix" (consistent with the input image size, the weight of each pixel is equal to the adjusted weight of the area to which it belongs); in the feature extraction layer of the CNN model (such as after the first convolution layer), this matrix is ​​multiplied point by point with the feature map to achieve dynamic suppression of features in areas with abnormal lighting.

[0044] If the brightness deviation is within the preset deviation standard range, it means that the brightness deviation within this range belongs to the normal fluctuation range and is not a feature distortion caused by abnormal lighting (such as extreme shadows or excessive highlights). In order to reduce invalid calculations and resource consumption and improve processing efficiency, the regional feature weights will not be dynamically adjusted. The preset deviation standard range represents the closed interval formed by the preset shadow brightness lower limit and the preset highlight brightness lower limit.

[0045] If the brightness deviation is within the preset shadow brightness range, it is recorded as a first-level shadow area. The accuracy assessment correction amount and the brightness deviation are weighted averaged and then rounded to negative infinity as the shadow area weight reduction amount, which can effectively balance the region's own characteristics and the system evaluation error. By rounding to negative infinity, the weight reduction force can be strengthened to ensure that the weight of the first-level shadow area can accurately match its actual impact in the image - it will not excessively interfere with normal feature extraction due to excessive weight, nor will it ignore its due role due to too low weight, thereby providing a more reliable feature basis for subsequent image analysis (such as target detection, scene understanding, etc.). The result of multiplying the preset shadow area weight and the shadow area weight reduction amount is used as the current shadow area weight to enhance the feature distinction between the first-level shadow area and other areas, and optimize the overall image feature distribution. The accuracy assessment correction amount represents the difference between the accuracy assessment threshold and the dynamic feature extraction accuracy assessment. The preset shadow brightness range represents the open interval formed by the preset shadow brightness lower limit and the preset shadow brightness upper limit.

[0046] If the brightness deviation is greater than or equal to the preset shadow brightness upper limit, it is recorded as a secondary shadow area. The result of the weighted average of the accuracy assessment correction amount and the deviation compensation amount and then rounded to positive infinity is used as the shadow area weight increase amount. The preset shadow area weight and the shadow area weight increase amount are multiplied as the current shadow area weight. The upward rounding strengthens the increase force, accurately improves the secondary shadow area weight to match its significant feature influence, dynamically associates the system state and regional features to enhance the adaptive adjustment capability, clarifies the calculation logic to improve interpretability and controllability, and at the same time strengthens the feature distinction from other areas and optimizes the overall feature distribution, thereby providing a more reliable feature basis for subsequent image analysis and improving processing accuracy. The deviation compensation amount represents the difference between the brightness deviation and the preset shadow brightness upper limit.

[0047] In this embodiment, through the coordinated operation of multiple links, accurate processing of image illumination abnormalities is achieved, bringing significant technical effects in multiple dimensions. The method of directly adjusting the highlights in non-critical areas according to the threshold and separately processing the eye highlights in combination with pupil detection reflects the flexibility and pertinence of the processing - it not only simplifies the processing flow of non-critical areas to improve efficiency, but also avoids the misjudgment of eye highlights (such as pupil reflections) through special processing, ensuring the processing accuracy of key feature areas. Overall, through standardized parameter definition, refined area processing, differentiated weight adjustment and deep integration with model feature extraction, it not only ensures the accuracy and pertinence of the processing of illumination abnormalities, but also improves the system's operating efficiency, adaptability and interpretability, providing a more reliable feature basis for subsequent tasks such as image target detection and scene understanding, effectively reducing misjudgments and missed judgments caused by illumination problems, and significantly enhancing the robustness of the image processing system in complex lighting environments.

[0048] As a further specific explanation, determining whether to dynamically adjust the regional feature weights also includes: If the brightness deviation is within the preset highlight brightness range, it is recorded as a first-level highlight area. The result of taking the arithmetic average of the accuracy assessment correction amount and the brightness deviation and rounding it down is used as the highlight area weight reduction amount. The result of multiplying the preset highlight area weight by the highlight area weight reduction amount is used as the current highlight area weight. The downward adjustment force is strengthened by rounding down to accurately match the actual impact of the first-level highlight area, avoiding excessive interference with normal feature extraction while retaining necessary weights, optimizing the overall feature distribution, improving the accuracy and pertinence of the system's highlight area processing, and providing a reliable feature basis for subsequent image analysis. The preset highlight brightness range represents the open interval formed by the preset highlight brightness lower limit and the preset highlight brightness upper limit.

[0049] If the brightness deviation is greater than or equal to the preset upper limit of highlight brightness, it is recorded as a secondary highlight area, and the result of taking the arithmetic average of the accuracy assessment correction amount and the deviation adjustment amount and rounding up is used as the highlight area weight increase amount, and the result of multiplying the preset highlight area weight by the highlight area weight increase amount is used as the current highlight area weight. Dynamic adjustment is achieved through the increase amount, which not only retains the empirical basic settings reflected by the preset weights, but also can be flexibly corrected according to the actual lighting conditions and system evaluation status. It can not only reflect the importance of such areas under general circumstances, but also perform real-time adaptation to the highlight intensity and system processing accuracy of specific scenes to ensure that the weight value can be accurate It reflects the actual feature importance of the secondary highlight area in the current image. Through precise area definition, multi-parameter fusion adjustment logic and enhanced rounding operation, it can dynamically optimize the weight according to the actual situation of the secondary highlight area, effectively offset the interference of strong light on feature extraction, and enable subsequent image analysis (such as target detection, feature matching, etc.) to more clearly capture the effective information of the area, optimize the feature distribution of the overall image, provide a more reliable feature basis for various image processing tasks, reduce feature loss or misjudgment caused by strong light, and improve the robustness of the system in complex lighting environments. The deviation adjustment amount represents the difference between the brightness deviation and the preset highlight brightness.

[0050] In this embodiment, precise suppression of highlight interference and feature optimization are achieved by meticulously demarcating primary and secondary highlight regions and implementing a differentiated weight adjustment strategy. Through precise region definition, multi-parameter fusion adjustment logic, and directional rounding (primary downward enhanced suppression, secondary upward dynamic adaptation), the interference of highlights of varying intensities on feature extraction is effectively offset. This not only avoids excessive influence from primary highlight regions, but also ensures effective information retention in secondary highlight regions under strong light conditions. This optimizes the overall image feature distribution, providing a reliable feature foundation for subsequent analysis tasks such as target detection and feature matching, reduces feature loss or misjudgment due to illumination, and significantly improves the system's processing accuracy in complex lighting environments.

[0051] like Figure 2 As shown, a flow chart of dynamic adjustment of facial area mask weights of a risk control identity authentication system based on facial expression recognition provided by an embodiment of the present application is provided. The specific logic is: whether to perform dynamic adjustment of facial area mask weights is determined according to mask entropy; if the mask entropy is within a preset mask entropy standard range, dynamic adjustment of facial area mask weights is not performed; if the mask entropy is less than a preset mask entropy lower bound, the accuracy assessment correction amount and the mask entropy compensation amount are harmonized and averaged, and the result of rounding to zero is used as the facial area mask weight increase amount; if the mask entropy is greater than the preset mask entropy upper bound, the accuracy assessment correction amount and the mask entropy correction amount are harmonized and averaged, and the result of rounding to zero is used as the facial area mask weight decrease amount.

[0052] To further clarify, the specific steps for determining whether to dynamically adjust facial region mask weights are as follows: Facial region mask weights are essentially a weight matrix or mask map with the same spatial dimensions (e.g., height × width) as the input facial image. The value of each pixel or region represents the importance of that region in the model's decision-making process. For example, a weight of 1 indicates that the region fully participates in the calculation; a weight of 0 indicates that the region is ignored; intermediate values ​​(e.g., 0.5) indicate partial participation or reduced influence. Mask entropy is a quantitative metric used to measure the degree of dispersion of information in the distribution of facial region mask weights.

[0053] If the mask entropy is within the preset mask entropy standard range, it means that the uncertainty in determining the current facial area mask is moderate, the information distribution is reasonable, and no dynamic weight adjustment is required, then the facial area mask weight is not dynamically adjusted. The preset mask entropy standard range represents a closed interval formed by the preset mask entropy lower bound and the preset mask entropy upper bound.

[0054] If the mask entropy is less than the preset mask entropy lower bound, it means that the uncertainty of the current facial region mask is too low, the information is too concentrated, and there is a risk of insufficient feature expression ability or oversensitivity to local details. The accuracy assessment correction amount and the mask entropy compensation amount are harmonically averaged and rounded to zero, and the result is used as the facial region mask weight increase amount. The facial region mask weight increase amount is multiplied by the preset facial region mask weight as the current facial region mask weight. Rounding to zero enhances the smoothness and conservatism of the weight increase, avoiding feature distribution imbalance caused by excessive adjustment. After the weight increase, the importance of the facial region in feature extraction is effectively enhanced, and the feature sparsity or local overfitting problems caused by low mask entropy are compensated, thereby improving the representation of facial details. By dynamically responding to abnormally low mask entropy states, refined compensation optimization of feature weights is achieved, ensuring rich feature expression in low-uncertainty scenarios, providing a more reliable feature foundation for subsequent face recognition, expression analysis and other tasks. The mask entropy compensation amount represents the difference between the preset mask entropy lower bound and the mask entropy.

[0055] If the mask entropy is greater than the preset mask entropy upper bound, it means that the uncertainty of the current facial area mask is too high, the information distribution is too dispersed, and noise or interference features may be introduced, affecting the effective expression of the facial area. The accuracy evaluation correction amount and the mask entropy correction amount are harmonically averaged and rounded to zero, and the result is used as the facial area mask weight reduction amount. The facial area mask weight reduction amount is multiplied by the preset facial area mask weight as the current facial area mask weight. The harmonic average combines the correction requirements of the system evaluation accuracy and the mask entropy exceeding the limit, and rounding to zero ensures the smoothness and conservatism of the weight reduction, avoiding excessive suppression. This results in the loss of key features. After the weight is lowered, the interference of high-uncertainty masks on feature extraction is effectively reduced, and the excessive influence of noise or redundant information is suppressed, thereby improving the purity and stability of facial area features. By dynamically responding to the abnormally high state of mask entropy, the refined suppression optimization of feature weights is achieved, ensuring that key features can still be reliably extracted in high-uncertainty scenarios, providing a more robust feature basis for subsequent face recognition, expression analysis and other tasks, and at the same time optimizing the system's adaptability and anti-interference ability under complex or low-quality mask conditions. The mask entropy correction amount represents the difference between the mask entropy and the preset mask entropy upper bound.

[0056] In this embodiment, a dynamic weight adjustment mechanism based on mask entropy achieves intelligent response and refined optimization to facial region mask uncertainty, significantly improving the robustness and adaptability of feature extraction. This overall mechanism dynamically responds to abnormal states of mask entropy to achieve refined compensation or suppression optimization of feature weights. This avoids feature distortion caused by over-adjustment while ensuring rich feature expression and anti-interference capabilities in both low and high uncertainty scenarios. This provides a more stable and reliable feature foundation for subsequent tasks such as face recognition and expression analysis, significantly improving the system's processing accuracy under complex or low-quality mask conditions.

[0057] Furthermore, the specific judgment steps for determining whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality evaluation are as follows: if the multi-dimensional comparison rationality evaluation is greater than or equal to the preset rationality evaluation upper limit, the multi-dimensional comparison rationality optimization is not performed, and a qualified comparison instruction is sent; if the multi-dimensional comparison rationality evaluation is within the preset rationality interval, whether to perform dynamic adjustment of the transition frame weight is determined based on the expression switching interval. If so, whether to perform dynamic adjustment of the static-dynamic feature weight is determined after the dynamic adjustment. Otherwise, whether to perform dynamic adjustment of the static-dynamic feature weight is directly determined. The preset rationality interval represents the open interval formed by the preset rationality evaluation lower limit and the preset rationality evaluation upper limit; if the multi-dimensional comparison rationality evaluation is less than or equal to the preset rationality evaluation lower limit, a qualified comparison instruction is sent to the preset reviewer.

[0058] like Figure 3 As shown, it is a whole flow chart of the dynamic adjustment of the transition frame weight of a risk control identity authentication system based on facial expression recognition provided by an embodiment of the present application. The specific logic is: whether to perform dynamic adjustment of the transition frame weight is determined according to the expression switching interval; if the expression switching interval is greater than the preset upper limit of the expression switching interval, the multi-dimensional comparison rationality evaluation and the switching interval compensation amount are input into the identity authentication database for mapping to obtain the transition frame weight increase amount; if the expression switching interval is within the preset expression switching interval range, the transition frame weight is not dynamically adjusted; if the expression switching interval is less than the preset lower limit of the expression switching interval, the multi-dimensional comparison rationality evaluation and the switching interval correction amount are input into the identity authentication database for mapping to obtain the transition frame weight decrease amount.

[0059] As a further specific explanation, the specific process of determining whether to dynamically adjust the transition frame weight according to the expression switching interval is as follows: the transition frame refers to all intermediate frames between the starting frame (initial expression) and the target frame (final expression) during the process of the expression switching from one state to another.

[0060] If the expression switching interval is greater than the preset upper limit, the multi-dimensional comparison rationality assessment and switching interval compensation are input into the identity authentication database to obtain the transition frame weight increase. The result of multiplying the transition frame weight increase by the preset transition frame weight is used as the current transition frame weight. The increased transition frame weight effectively enhances the importance of transition frames under long expression states, compensating for the feature sparsity or weakened identity information caused by excessive switching intervals, thereby improving the contribution of transition frames in identity authentication or expression analysis. By dynamically responding to abnormally extended expression switching intervals, refined compensation optimization of transition frame weights is achieved, ensuring the continuity and reliability of identity features in scenarios with slow or stagnant expression switching, providing more robust feature support for subsequent identity authentication, expression recognition, and other tasks. It also enhances the system's adaptability and accuracy under complex or non-ideal expression switching conditions. The switching interval compensation represents the difference between the expression switching interval and the preset upper limit of the expression switching interval.

[0061] If the expression switching interval is within the preset expression switching interval range, it means that the current expression state switching frequency is reasonable, the information distribution of the transition frame is continuous and sufficient, and there is no need to dynamically optimize the transition frame weight. The preset expression switching interval represents the closed interval formed by the preset expression switching interval lower limit and the preset expression switching interval upper limit.

[0062] If the expression switching interval is less than the preset expression switching interval lower limit, it means that the current expression state is judged to switch too frequently, and the transition frame may have information redundancy or noise interference, which affects the accuracy and stability of the identity feature expression. The multi-dimensional comparison rationality evaluation and the switching interval correction amount are input into the identity authentication database mapping to obtain the transition frame weight reduction amount. The result of multiplying the transition frame weight reduction amount by the preset transition frame weight is used as the current transition frame weight. Using the identity authentication database mapping relationship, the abnormal degree of expression switching frequency and the multi-dimensional identity feature comparison results are comprehensively considered to ensure that the weight reduction amount reflects both the degree of deviation of the switching interval and the rationality of the identity authentication, thereby achieving precise suppression. By lowering the weight of transition frames, the system effectively reduces the interference of transition frames in frequent switching states, suppresses the impact of redundant information and noise on identity feature extraction, and thus enhances the dominant role of key frames in identity authentication or expression analysis. By dynamically responding to the abnormal shortening of the expression switching interval, the system achieves refined suppression and optimization of the transition frame weight, ensuring that the purity and reliability of identity features can be maintained in scenarios where expression switching is too frequent, providing more robust feature support for subsequent identity authentication, expression recognition and other tasks, and at the same time enhancing the system's adaptability and anti-interference ability under conditions of rapid switching or non-ideal expression changes. The switching interval correction value represents the difference between the preset lower limit of the expression switching interval and the expression switching interval.

[0063] In this embodiment, a dynamic weight adjustment mechanism based on the expression switching interval achieves intelligent response and refined optimization of the importance of transition frames, significantly improving the accuracy of identity authentication and expression analysis under complex expression changes. By dynamically responding to the different abnormal states of the expression switching interval, the overall mechanism achieves refined compensation or suppression optimization of transition frame weights. This not only avoids feature distortion caused by over-adjustment, but also ensures robust and reliable feature support in all expression switching scenarios, laying a highly efficient technical foundation for subsequent identity authentication, expression recognition, and other tasks.

[0064] To further clarify, the specific steps for determining whether to dynamically adjust static and dynamic feature weights are as follows: Static feature weights are used to quantify the contribution of static facial features in the rationality assessment of multi-dimensional comparisons. Static facial features refer to relatively stable, inherent features of the face, such as the geometric proportions of the facial contour (such as the distance between eyes and the ratio of nose width to face width), skin texture (the distribution of freckles and wrinkles), and the coordinate positions of key feature points (such as the fixed relative positions of the corners of the eyes, nose tip, and mouth corners). These features are not easily altered by changes in expression and are the core basis for identity recognition. Dynamic feature weights are used to measure the importance of dynamic facial features in the rationality assessment of multi-dimensional comparisons. Dynamic facial features refer to motion features that arise with changes in expression, such as the speed of expression switching, the trajectory of muscle movement in transition frames (such as the amplitude of the mouth corner's upward movement or the rate of brow contraction), and the time interval between transitions between different expression states. These features reflect the naturalness and consistency of expression changes and can be used to distinguish authentic expressions from fabricated ones (such as those synthesized from photos or videos).

[0065] If the positioning error of a single feature point is less than the asymptotic lower bound of the preset positioning error, it means that the current feature point positioning accuracy is extremely high and the static feature information is extremely reliable, but there may be a risk of relatively insufficient or redundant dynamic feature expression. In this case, the multi-dimensional comparison rationality assessment and the positioning error comparison amount are input into the identity authentication database to map to obtain a static feature weight increase instruction. The static feature weight increase instruction represents the result of arithmetic averaging the rationality assessment and the positioning error comparison amount. The static feature weight increase instruction is input into the identity authentication database to map to obtain a dynamic feature decrease instruction. This comprehensively considers the accuracy of feature point positioning and the multi-dimensional identity feature comparison results to ensure that the static feature weight increase reflects both the excellent performance of the positioning error and the rationality of identity authentication, thereby achieving precise enhancement. By increasing the static feature weight and simultaneously decreasing the dynamic feature weight, the system effectively strengthens the dominant role of high-precision static features in identity authentication, suppresses possible redundancy or noise interference in dynamic features, and thus improves the purity and stability of the overall feature expression. By dynamically responding to the abnormally excellent state of feature point positioning error, a refined balance optimization of static and dynamic feature weights is achieved, ensuring that identity features can be reliably expressed even in scenarios where feature point positioning is extremely precise, providing more robust feature support for subsequent identity authentication, feature matching and other tasks, while enhancing the system's adaptability and anti-interference capabilities under high-quality feature point positioning conditions. The positioning error comparison value represents the difference between the preset asymptotic lower bound of the positioning error and the positioning error of a single feature point.

[0066] If the positioning error of a single feature point is within the preset positioning error interval, it is determined that the current feature point positioning accuracy is within a reasonable range, the information distribution between static features and dynamic features is balanced and complementary, and there is no need to dynamically adjust the static-dynamic feature weights. The preset positioning error interval represents the closed interval formed by the asymptotic lower bound of the preset positioning error and the asymptotic upper bound of the preset positioning error.

[0067] If the positioning error of a single feature point is greater than the asymptotic upper bound of the preset positioning error, it means that the current feature point positioning accuracy is insufficient and the static feature information is unreliable, which may affect the accuracy of identity authentication or feature matching. The multi-dimensional comparison rationality evaluation and the positioning error correction amount are input into the identity authentication database for mapping to obtain a static feature weight reduction instruction. The static feature weight reduction instruction represents the result of arithmetic averaging the rationality evaluation and the positioning error correction amount. The static feature weight reduction instruction is input into the identity authentication database for mapping to obtain a dynamic feature increase instruction. By lowering the static feature weight and simultaneously increasing the dynamic feature weight, the risk of misjudgment introduced by the excessive static feature positioning error is effectively reduced, and at the same time, the compensatory role of dynamic features in identity authentication is enhanced, thereby improving the adaptability and robustness of the overall feature expression. By dynamically responding to the abnormally high state of feature point positioning error, a refined balance optimization of static and dynamic feature weights is achieved, ensuring that the reliable expression and stability of identity features can be maintained even in scenarios with poor feature point positioning quality, providing more robust feature support for subsequent identity authentication, feature matching and other tasks, while enhancing the system's adaptability and anti-interference ability under low-quality feature point positioning conditions. The positioning error correction value represents the difference between the positioning error of a single feature point and the asymptotic upper bound of the preset positioning error.

[0068] In this embodiment, through a dynamic weight adjustment mechanism based on the positioning error of a single feature point, intelligent response and refined balance optimization of static and dynamic feature weights are achieved, thereby significantly improving the accuracy of identity authentication and feature matching. The positioning error and the rationality of multi-dimensional comparison are comprehensively considered through the database mapping relationship, and refined regulation of feature weights in different positioning accuracy scenarios is achieved. This not only avoids feature distortion caused by excessive adjustment, but also ensures that the reliable expression and robustness of identity features can be maintained under conditions of extremely high, moderate or poor feature point positioning, providing more robust feature support for subsequent identity authentication, feature matching and other tasks, while significantly enhancing the system's adaptability and anti-interference ability under complex or low-quality feature point positioning conditions.

[0069] like Figure 4As shown, a flow chart of a risk control identity authentication method based on facial expression recognition provided by an embodiment of the present application, a risk control identity authentication method based on facial expression recognition provided by an embodiment of the present application includes: collecting a facial image sequence of a user to be authenticated; preprocessing the facial image sequence to obtain feature extraction parameters to quantify the accuracy of CNN and LSTM models in the facial expression feature extraction process, and obtaining a dynamic feature extraction accuracy evaluation; judging whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy evaluation, and if so, performing a facial expression feature comparison link after optimization, otherwise directly performing a facial expression feature comparison link, and the dynamic feature extraction accuracy optimization includes dynamic adjustment of regional feature weights and dynamic adjustment of facial region mask weights; obtaining multi-dimensional comparison parameters in the facial expression feature comparison link to quantify the rationality of the facial expression feature comparison process, and obtaining a multi-dimensional comparison rationality evaluation; judging whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality evaluation, and if so, generating a risk control identity authentication decision after optimization, otherwise directly generating a risk control identity authentication decision, and the multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights and dynamic adjustment of static-dynamic feature weights.

[0070] As an embodiment of the second aspect, a user logs in through a mobile banking app and adopts a dynamic identity authentication mechanism based on a facial expression sequence. During the login process, the user is required to complete a facial expression action sequence of smiling → natural → blinking. By collecting their facial image sequence and combining CNN and LSTM models for dynamic feature extraction and comparison, a risk control identity authentication decision is finally generated. The image sequence is then preprocessed with face detection, alignment, normalization, and illumination compensation to generate standardized facial region input, which is then input into a CNN-LSTM hybrid model for feature extraction. CNN extracts spatial features (such as facial contours and texture), and LSTM models temporal dynamic features (such as expression change trajectories). At the same time, the accuracy of dynamic feature extraction is evaluated by calculating the cross-entropy loss and cosine similarity of the model output and the annotated expression label. The extracted dynamic feature sequence is compared with the user registration template, and the system performs multi-dimensional comparison rationality optimization. Based on the final comparison result, the system outputs "authentication passed" to allow login. If an abnormality is detected, secondary verification is triggered or high-risk operations are restricted. Through dynamic weight adjustment and multi-dimensional optimization mechanisms, the accuracy, robustness, and security of identity authentication can be maintained in scenarios such as complex lighting, blurred expressions, or abnormal switching.

[0071] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A risk control identity authentication system based on facial expression recognition, characterized in that: It includes identity information collection module, facial feature extraction module, dynamic feature extraction accuracy analysis module, multi-dimensional comparison module and multi-dimensional comparison rationality analysis module: Wherein, the identity information acquisition module is used to acquire a facial image sequence of the user to be authenticated; The facial feature extraction module is used to preprocess the facial image sequence to obtain feature extraction parameters to quantify the accuracy of the CNN and LSTM models in the facial expression feature extraction process and obtain a dynamic feature extraction accuracy assessment; The dynamic feature extraction accuracy analysis module is used to determine whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy assessment. If so, a facial expression feature comparison link is performed after optimization; otherwise, the facial expression feature comparison link is directly performed. The dynamic feature extraction accuracy optimization includes dynamic adjustment of regional feature weights and dynamic adjustment of facial region mask weights. The multidimensional comparison module is used to obtain multidimensional comparison parameters in the facial expression feature comparison link to quantify the rationality of the facial expression feature comparison process and obtain a multidimensional comparison rationality assessment; The multi-dimensional comparison rationality analysis module is used to determine whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality evaluation. If so, a risk control identity authentication decision is generated after optimization; otherwise, a risk control identity authentication decision is directly generated. The multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights and dynamic adjustment of static-dynamic feature weights.

2. A risk control identity authentication system based on facial expression recognition according to claim 1, characterized in that: The feature extraction parameters include: key frame missing rate, single feature point displacement vector field and frame interval time; Feature extraction parameter thresholds and feature extraction parameter correction factors are obtained from the constructed identity authentication database. The feature extraction parameter thresholds include a key frame missing rate threshold, a single feature point displacement vector field threshold, and a frame interval time standard value. The feature extraction parameter correction factors include a key frame missing rate correction factor, a single feature point displacement vector field correction factor, and a frame interval time correction factor.

3. A risk control identity authentication system based on facial expression recognition according to claim 1, characterized in that: The specific steps for determining whether to perform dynamic feature extraction accuracy optimization are as follows: Obtaining an accuracy assessment threshold from the constructed identity authentication database; If the dynamic feature extraction accuracy evaluation is not less than the accuracy evaluation threshold, the dynamic feature extraction accuracy optimization is not performed, and a feature extraction accuracy qualified instruction is sent to the multi-dimensional comparison module; If the dynamic feature extraction accuracy assessment is less than the accuracy assessment threshold, the brightness deviation is used to determine whether to dynamically adjust the regional feature weight. If so, after the dynamic adjustment, it is determined whether to dynamically adjust the facial region mask weight. Otherwise, it is directly determined whether to dynamically adjust the facial region mask weight.

4. A risk control identity authentication system based on facial expression recognition according to claim 3, characterized in that: The specific process of determining whether to dynamically adjust the regional feature weight according to the brightness deviation is as follows: Obtaining a preset shadow brightness upper limit, a preset shadow brightness lower limit, a preset highlight brightness lower limit, and a preset highlight brightness upper limit from a constructed identity authentication database, wherein the preset shadow brightness upper limit is less than the preset shadow brightness lower limit, the preset shadow brightness lower limit is less than the preset highlight brightness lower limit, and the preset highlight brightness lower limit is less than the preset highlight brightness upper limit; If the brightness deviation is within the preset deviation standard interval, the regional feature weight is not dynamically adjusted. The preset deviation standard interval represents a closed interval formed by the preset shadow brightness lower limit and the preset highlight brightness lower limit. If the brightness deviation is within the preset shadow brightness range, it is recorded as a first-level shadow area. The accuracy assessment correction amount and the brightness deviation are weighted averaged and then rounded to negative infinity as the shadow area weight reduction amount. The preset shadow area weight and the shadow area weight reduction amount are multiplied together as the current shadow area weight. The accuracy assessment correction amount is used to represent the deviation between the accuracy assessment threshold and the dynamic feature extraction accuracy assessment. The preset shadow brightness range represents an open interval formed by a preset shadow brightness lower limit and a preset shadow brightness upper limit. If the brightness deviation is greater than or equal to the preset shadow brightness upper limit, it is recorded as a secondary shadow area. The result of weighted averaging the accuracy assessment correction amount and the deviation compensation amount and rounding to positive infinity is used as the shadow area weight increase amount. The result of multiplying the preset shadow area weight and the shadow area weight increase amount is used as the current shadow area weight. The deviation compensation amount is used to characterize the deviation between the brightness deviation and the preset shadow brightness upper limit.

5. A risk control identity authentication system based on facial expression recognition according to claim 4, characterized in that: The determining whether to perform dynamic adjustment of regional feature weights further includes: If the brightness deviation is within the preset highlight brightness range, it is recorded as a first-level highlight area. The accuracy assessment correction amount and the brightness deviation are arithmetic averaged and rounded down as the highlight area weight reduction amount. The preset highlight area weight and the highlight area weight reduction amount are multiplied together as the current highlight area weight. The preset highlight brightness range represents an open interval formed by a preset highlight brightness lower limit and a preset highlight brightness upper limit. If the brightness deviation is greater than or equal to the preset upper limit of highlight brightness, it is recorded as a secondary highlight area. The result of taking the arithmetic average of the accuracy assessment correction amount and the deviation adjustment amount and rounding it up is used as the highlight area weight increase amount. The result of multiplying the preset highlight area weight and the highlight area weight increase amount is used as the current highlight area weight. The deviation adjustment amount is used to characterize the deviation between the brightness deviation and the preset highlight brightness.

6. A risk control identity authentication system based on facial expression recognition according to claim 3, characterized in that: The specific steps of determining whether to perform dynamic adjustment of the facial area mask weight are: If the mask entropy is within a preset mask entropy standard interval, no dynamic adjustment of the facial region mask weight is performed, wherein the preset mask entropy standard interval represents a closed interval formed by a preset mask entropy lower bound and a preset mask entropy upper bound; If the mask entropy is less than the preset mask entropy lower bound, the accuracy assessment correction amount and the mask entropy compensation amount are harmonically averaged and rounded to zero, and the result is used as the facial region mask weight increase amount. The facial region mask weight increase amount is multiplied by the preset facial region mask weight as the current facial region mask weight. The mask entropy compensation amount is used to represent the deviation between the preset mask entropy lower bound and the mask entropy. If the mask entropy is greater than the preset mask entropy upper bound, the result of harmonically averaging the accuracy assessment correction amount and the mask entropy correction amount and rounding to zero is used as the facial area mask weight reduction amount, and the result of multiplying the facial area mask weight reduction amount by the preset facial area mask weight is used as the current facial area mask weight. The mask entropy correction amount is used to characterize the deviation between the mask entropy and the preset mask entropy upper bound.

7. The risk control identity authentication system based on facial expression recognition according to claim 1, characterized in that: The multi-dimensional comparison parameters include: the accuracy evaluation of the dynamic feature extraction to be compared, the positioning error of a single feature point and the expression switching interval; Obtaining a multi-dimensional comparison parameter threshold and a multi-dimensional comparison parameter correction factor from the constructed identity authentication database, wherein the multi-dimensional comparison parameter threshold includes a dynamic feature extraction accuracy assessment threshold, a single feature point positioning error standard value, and an expression switching interval standard value; and the multi-dimensional comparison parameter correction factor includes a dynamic feature extraction accuracy assessment correction factor, a single feature point positioning error correction factor, and an expression switching interval correction factor; The specific steps for judging whether to perform multi-dimensional comparison rationality optimization based on the multi-dimensional comparison rationality evaluation are as follows: If the multi-dimensional comparison rationality evaluation is greater than or equal to the preset rationality evaluation upper limit, the multi-dimensional comparison rationality optimization is not performed, and a qualified comparison instruction is sent; If the rationality evaluation of the multi-dimensional comparison is within the preset rationality interval, it is determined whether to perform dynamic adjustment of the transition frame weight according to the expression switching interval. If so, it is determined whether to perform dynamic adjustment of the static-dynamic feature weight after the dynamic adjustment. Otherwise, it is directly determined whether to perform dynamic adjustment of the static-dynamic feature weight. The preset rationality interval represents an open interval formed by the preset rationality evaluation lower bound and the preset rationality evaluation upper bound; If the rationality assessment of the multi-dimensional comparison is less than or equal to the preset rationality assessment lower limit, a comparison failure instruction is sent to the preset reviewer.

8. The risk control identity authentication system based on facial expression recognition according to claim 7, characterized in that: The specific process of determining whether to dynamically adjust the transition frame weight according to the expression switching interval is as follows: If the expression switching interval is greater than the preset expression switching interval upper limit, the multi-dimensional comparison rationality evaluation and the switching interval compensation amount are input into the identity authentication database for mapping to obtain a transition frame weight increase amount, and the result of multiplying the transition frame weight increase amount by the preset transition frame weight is used as the current transition frame weight. The switching interval compensation amount is used to represent the deviation between the expression switching interval and the preset expression switching interval upper limit; If the expression switching interval is within a preset expression switching interval, no transition frame weight optimization is performed, wherein the preset expression switching interval represents a closed interval formed by a preset expression switching interval lower limit and a preset expression switching interval upper limit; If the expression switching interval is less than the preset expression switching interval lower limit, the multi-dimensional comparison rationality evaluation and the switching interval correction amount are input into the identity authentication database for mapping to obtain the transition frame weight reduction amount, and the result of multiplying the transition frame weight reduction amount by the preset transition frame weight is used as the current transition frame weight. The switching interval correction amount is used to characterize the deviation between the preset expression switching interval lower limit and the expression switching interval.

9. The risk control identity authentication system based on facial expression recognition according to claim 7, characterized in that: The specific steps of determining whether to perform dynamic adjustment of static-dynamic feature weights are as follows: If the positioning error of a single feature point is less than the preset asymptotic lower bound of the positioning error, the multi-dimensional comparison rationality evaluation and the positioning error control amount are input into the identity authentication database for mapping to obtain a static feature weight increase instruction, wherein the static feature weight increase instruction represents the result obtained by arithmetic averaging the rationality evaluation and the positioning error control amount, and the static feature weight increase instruction is input into the identity authentication database for mapping to obtain a dynamic feature decrease instruction, wherein the positioning error control amount is used to represent the deviation between the preset asymptotic lower bound of the positioning error and the positioning error of a single feature point; If the positioning error of a single feature point is within a preset positioning error interval, no dynamic adjustment of the static-dynamic feature weight is performed, wherein the preset positioning error interval represents a closed interval formed by the asymptotic lower bound of the preset positioning error and the asymptotic upper bound of the preset positioning error; If the positioning error of a single feature point is greater than the preset asymptotic upper bound of the positioning error, the multi-dimensional comparison rationality evaluation and the positioning error correction amount are input into the identity authentication database for mapping to obtain a static feature weight reduction instruction. The static feature weight reduction instruction represents the result obtained by arithmetic averaging the rationality evaluation and the positioning error correction amount. The static feature weight reduction instruction is input into the identity authentication database for mapping to obtain a dynamic feature increase instruction. The positioning error correction amount is used to characterize the deviation between the positioning error of a single feature point and the preset asymptotic upper bound of the positioning error.

10. A risk control identity authentication method based on facial expression recognition, wherein the risk control identity authentication method based on facial expression recognition is applied to a risk control identity authentication system based on facial expression recognition as described in any one of claims 1 to 9, characterized in that: include: Collecting a facial image sequence of the user to be authenticated; Preprocess the facial image sequences to obtain feature extraction parameters to quantify the accuracy of CNN and LSTM models in facial expression feature extraction and obtain dynamic feature extraction accuracy evaluation; Determine whether to perform dynamic feature extraction accuracy optimization based on the dynamic feature extraction accuracy evaluation; if so, perform a facial expression feature comparison step after the optimization; otherwise, perform the facial expression feature comparison step directly; the dynamic feature extraction accuracy optimization includes dynamic adjustment of regional feature weights and dynamic adjustment of facial region mask weights; Obtaining multi-dimensional comparison parameters in the facial expression feature comparison process to quantify the rationality of the facial expression feature comparison process and obtain a multi-dimensional comparison rationality evaluation; Based on the multi-dimensional comparison rationality assessment, it is determined whether to perform multi-dimensional comparison rationality optimization. If so, the risk control identity authentication decision is generated after optimization; otherwise, the risk control identity authentication decision is directly generated. The multi-dimensional comparison rationality optimization includes dynamic adjustment of transition frame weights and dynamic adjustment of static-dynamic feature weights.

Images