Face image quality evaluation method and system and computer readable storage medium

Abstract

The invention discloses a human face image quality assessment method and system and a computer readable storage medium, and relates to the technical field of human face image quality assessment, and the method specifically comprises the following steps: uniformly dividing a human face image captured by a camera in real time into a plurality of regions through a face key point detection method under the condition that there is an artifact effect; acquiring artifact quality evaluation information of each region of the face image in real time, evaluating the quality influence degree of artifacts on each region, and dividing each region into a high influence region, a middle influence region and a low influence region according to an evaluation result; and according to the division result of each region of the face image, constructing a dynamic intervention mechanism, and respectively performing corresponding intervention measures on the high-influence region, the medium-influence region and the low-influence region. According to the method, the problem of human face image quality evaluation misjudgment caused by artifact interference in a dynamic light source environment is solved, and accurate evaluation and efficient dynamic optimization of the image quality are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of face image quality assessment, and in particular to a face image quality assessment method, system and computer-readable storage medium. Background Art

[0002] A face image refers to a digital image containing human facial features acquired through an image acquisition device (such as a camera, webcam, mobile phone, etc.). These images are usually used in applications such as face recognition, sentiment analysis, behavior monitoring, and security monitoring. Because face images are affected by multiple factors such as lighting, shooting angle, resolution, occlusion, and noise, they may be blurred, distorted, and have low contrast, which seriously affect the accuracy of subsequent face recognition and analysis. Therefore, it is particularly important to evaluate the quality of face images. Quality evaluation can help us quantify the clarity, contrast, noise level, and integrity of facial features of the image, providing more reliable and accurate input for subsequent tasks such as face recognition and sentiment analysis. By evaluating the quality of the image, we can promptly discover and correct defects in the image, thereby improving the stability and accuracy of the recognition system. In addition, quality evaluation can also provide feedback to the automated system, optimize the image acquisition process, and ensure that high-quality input data is ultimately obtained. Therefore, image quality evaluation is not only a key step in improving the effect of face image processing tasks, but also the basis for many computer vision systems to operate accurately and efficiently.

[0003] Existing face image quality assessment technologies usually combine traditional image processing methods and deep learning technologies to evaluate image quality in a variety of ways. Traditional methods judge image quality by analyzing low-level features of images, such as clarity, contrast, noise, brightness, etc. Specifically, clarity assessment usually uses edge detection algorithms (such as Sobel operator or Laplace operator) to analyze details in the image, noise assessment identifies noise levels by calculating the standard deviation or mean of the image, and contrast assessment determines the contrast of the image by calculating the brightness difference in the local area. At the same time, deep learning methods automatically learn high- and low-level features in the image by training convolutional neural networks (CNNs) and output a quality score, which can comprehensively consider all aspects of the image and make end-to-end quality predictions. In addition, generative adversarial networks (GANs) are also used to evaluate image quality, identifying quality problems in images by comparing the differences between generated high-quality images and original images. Based on these technologies, the evaluation process usually includes feature extraction, quality scoring and analysis of images, thereby providing reliable quality assurance for subsequent face recognition or analysis tasks. In general, these technologies can accurately quantify image quality problems and provide effective feedback for automated processing.

[0004] The prior art has the following deficiencies: In dynamic advertising display scenes in large shopping malls, cameras capture facial images in real time for user recognition, and dynamic light sources (such as neon lights or reflected light) will produce high reflected light in certain areas of the face (such as the forehead) and spread to adjacent areas (such as around the eyes), forming an artifact effect. This artifact is manifested as unnatural brightness transitions or patchy texture interference. This is because dynamic light reflection not only changes the light intensity in the highlight area, but also affects the texture features and brightness distribution of adjacent areas through light propagation, causing the real features of these areas to be covered or distorted. Since the intensity and position of the artifact change dynamically with the angle and brightness of the light source, the existing face image quality assessment technology relies on independent brightness, contrast or texture analysis of each area, and cannot accurately distinguish the difference between the artifact effect and the real image feature defects, resulting in confusion in the quality assessment results. In this case, the system may mistakenly judge the artifact area as quality abnormality and take unnecessary intervention measures (such as adjusting exposure or adding fill light), which will interfere with the overall image quality. At the same time, areas covered by artifact effects (such as periocular texture) may be misjudged as abnormal areas, reducing the accuracy of face recognition algorithm in extracting key features, ultimately affecting recognition efficiency and the accuracy of advertising placement decisions.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a facial image quality assessment method, system and computer-readable storage medium to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for evaluating the quality of a face image, specifically comprising the following steps: Under dynamic light source environment, the face image captured by the camera in real time is tested for artifact effects. By analyzing the uniformity of image brightness distribution, gradient direction anomalies, and texture feature changes, it is determined whether there is an artifact effect. When it is determined that there is an artifact effect, the face image captured by the camera in real time is evenly divided into several areas through a facial key point detection method; Acquire artifact quality assessment information of each area of ​​the face image in real time, analyze it after acquisition, assess the impact of artifacts on the quality of each area, and divide each area into high-impact area, medium-impact area and low-impact area according to the assessment results; According to the division results of each area of ​​the face image, a dynamic intervention mechanism is constructed to take corresponding intervention measures for high-impact areas, medium-impact areas and low-impact areas respectively; In the process of dynamic intervention mechanism intervening in various regions, the intervention feedback information of each region is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations, and optimize the dynamic intervention mechanism according to the evaluation results; Real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process are carried out to continuously improve artifact suppression methods and dynamic intervention strategies and optimize the facial image quality assessment process.

[0008] Preferably, the artifact quality assessment information of each area of ​​the face image is obtained in real time, and analyzed after acquisition to assess the impact of the artifact on the quality of each area, and the areas are divided into high impact areas, medium impact areas and low impact areas according to the assessment results. Specifically, the following steps are included: Acquire artifact quality assessment information of each area of ​​the face image in real time and perform preprocessing after acquisition; Extracting brightness distribution characteristic information and texture structure characteristic information from artifact quality assessment information of each region of the preprocessed face image, and analyzing them after extraction to generate brightness balance deviation index and detail texture destruction index of each region respectively; An artifact impact assessment model is constructed for the brightness balance deviation index and detail texture destruction index of each generated area, and the artifact impact coefficient of each area is generated. After the generation, the model is analyzed to evaluate the impact of the artifact on the quality of each area. According to the evaluation results, each area is divided into high-impact area, medium-impact area and low-impact area.

[0009] Preferably, the logic for obtaining the brightness balance deviation index and detail texture destruction index of each region is as follows: The brightness distribution characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average and variance of the brightness values ​​of all pixels in each area of ​​the face image at different times over a period of time, and the absolute average difference between the brightness values ​​of the edge pixels and the neighboring edge pixels of each area, and calibrated as , and , Indicates that within a period of time Moment face image The average brightness value of all pixels in the area Indicates that within a period of time Moment face image The variance of the brightness values ​​of all pixels in the region is Indicates that within a period of time Moment The absolute average difference between the brightness values ​​of the edge pixels in the region and the edge pixels in the neighborhood, , , and All are positive integers; Calculate the brightness balance deviation index of each area. The specific calculation formula is as follows:

[0010] In the formula, For the Brightness balance deviation index of each area; The texture structure characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average change rate, amplitude and directional distribution variance of the pixel gradient in each area of ​​the face image at different times over a period of time, and calibrated as , and , Indicates that within a period of time Moment face image The average rate of change of pixel gradients in a region is Indicates that within a period of time Moment face image The magnitude of the pixel gradient in the region is Indicates that within a period of time Moment face image The directional distribution variance of pixel gradients in a region; Calculate the detail texture destruction index of each area. The specific calculation formula is as follows:

[0011] In the formula, For the The detail texture destruction index of the area.

[0012] Preferably, the brightness balance deviation index of each region is generated and detail texture destruction index Construct an artifact impact assessment model and generate the artifact impact coefficient of each area through weighted summation , and the artifact influence coefficients of each region generated The threshold interval of the pre-set artifact influence coefficient Compare and evaluate the impact of artifacts on the quality of each area based on the comparison results, and divide each area into high-impact area, medium-impact area and low-impact area based on the evaluation results. The specific comparison analysis and division are as follows: like , if the artifact has a low impact on the quality of the area, then the area is classified as a low-impact area; like , if the artifact has a moderate impact on the quality of the area, then the area is classified as a medium-impact area; like If the artifact has a high degree of impact on the quality of the area, the area is divided into a high-impact area.

[0013] Preferably, a dynamic intervention mechanism is constructed according to the division results of each area of ​​the face image, specifically: according to the division results of the high-impact area, the medium-impact area and the low-impact area, different intervention measure parameters are set respectively to form a dynamic intervention mechanism; the dynamic intervention mechanism is based on the artifact influence coefficient of each area, and automatically determines the adjustment method and amplitude of brightness adjustment, texture enhancement and edge smoothing through pre-set rules; Corresponding intervention measures are taken for high-impact areas, medium-impact areas and low-impact areas respectively. Specifically, in high-impact areas, the artifact suppression enhancement parameters in the dynamic intervention mechanism are used to significantly adjust the brightness distribution, strengthen texture enhancement and optimize edge smoothing; in medium-impact areas, the artifact moderate adjustment parameters in the dynamic intervention mechanism are used to moderately optimize the brightness distribution and texture characteristics; in low-impact areas, the artifact stabilization parameters in the dynamic intervention mechanism are maintained without adjustment, and only the regional status is monitored to maintain stable quality.

[0014] Preferably, in the process of the dynamic intervention mechanism intervening in each area, the intervention feedback information of each area is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each area can achieve the expected effect, and optimize the dynamic intervention mechanism according to the evaluation results, which specifically includes the following steps: In the process of dynamic intervention mechanism intervening in each area, the intervention feedback information of each area is obtained in real time and preprocessed after acquisition; Extracting brightness adjustment feedback information and texture optimization feedback information from the preprocessed intervention feedback information of each region, and analyzing them after acquisition to generate a brightness adjustment deviation index and a texture restoration consistency index for each region respectively; An intervention effect evaluation model is constructed for the brightness adjustment deviation index and texture restoration consistency index generated for each area, and the intervention coefficient of each area is generated. The generated intervention coefficient of each area is compared with the pre-set intervention coefficient threshold of each area. According to the comparison results, it is evaluated whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and the dynamic intervention mechanism is optimized according to the evaluation results.

[0015] Preferably, the logic for obtaining the brightness adjustment deviation index and the texture restoration consistency index of each region is as follows: The brightness adjustment feedback information in the intervention feedback information of each area after preprocessing is extracted, specifically including the average value and variance of the pixel brightness values ​​of each area at different times within a period of time after the intervention, and the preset average value of the pixel brightness values ​​of each area after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average value of the pixel brightness in the region Indicates that within a period of time after the intervention Moment The variance of the pixel brightness values ​​in the region, The pre-specified intervention The average value of the pixel brightness in the region , , and All are positive integers; Calculate the brightness adjustment deviation index of each area. The specific calculation formula is as follows:

[0016] In the formula, For the Brightness adjustment deviation index for each region; The texture optimization feedback information in the intervention feedback information of each region after preprocessing is extracted, including the average amplitude of the pixel gradient of each region at different times within a period of time after the intervention, the directional distribution variance, and the pre-set average amplitude of the pixel gradient of each region after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average magnitude of the pixel gradient in the region, Indicates that within a period of time after the intervention Moment The directional distribution variance of the pixel gradient in the region, The pre-specified intervention The average amplitude of pixel gradients in a region; Calculate the texture restoration consistency index of each area. The specific calculation formula is as follows:

[0017] In the formula, For the Texture restoration consistency index for each region.

[0018] Preferably, the brightness of each region is adjusted to a deviation index and texture recovery consistency index Construct an intervention effect evaluation model and generate intervention coefficients for each region through weighted summation , and the intervention coefficients of each region generated The intervention coefficient thresholds for each area are set in advance Compare and evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations based on the comparison results, and optimize the dynamic intervention mechanism based on the evaluation results. The specific comparison and analysis are as follows: like , the intervention effect of the dynamic intervention mechanism in this area can achieve the expected effect, and there is no need to optimize the dynamic intervention mechanism; like The intervention effect of the dynamic intervention mechanism in this area cannot meet the expectations, and the dynamic intervention mechanism needs to be optimized, including: increasing the brightness adjustment intensity of the area and reducing the brightness distribution deviation; increasing the parameter weight of texture enhancement, strengthening the contrast of details and high-frequency texture characteristics; combining the real-time intervention feedback data of the area, dynamically adjusting the intervention parameters and rules, and optimizing the execution process of the intervention mechanism.

[0019] Preferably, the face image quality assessment system includes an artifact effect detection module, a region division and calibration module, an artifact quality assessment module, a dynamic intervention execution module, an intervention effect optimization module, and a comprehensive monitoring improvement module; The artifact effect detection module performs artifact effect detection on the face image captured by the camera in real time under a dynamic light source environment. It determines whether there is an artifact effect by analyzing the uniformity of the image's brightness distribution, abnormal gradient direction, and changes in texture features. The region division and calibration module evenly divides the face image captured by the camera in real time into several regions through the facial key point detection method when it is determined that there is an artifact effect; The artifact quality assessment module obtains the artifact quality assessment information of each area of ​​the face image in real time, analyzes it after acquisition, evaluates the impact of the artifact on the quality of each area, and divides each area into high-impact area, medium-impact area and low-impact area according to the assessment results; The dynamic intervention execution module builds a dynamic intervention mechanism based on the division results of each area of ​​the face image, and takes corresponding intervention measures for the high-impact area, medium-impact area and low-impact area respectively; The intervention effect optimization module obtains the intervention feedback information of each area in real time during the process of the dynamic intervention mechanism intervening in each area, analyzes it after obtaining, evaluates whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and optimizes the dynamic intervention mechanism according to the evaluation results; The comprehensive monitoring and improvement module conducts real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process, continuously improves the artifact suppression method and dynamic intervention strategy, and optimizes the facial image quality assessment process.

[0020] Preferably, a computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a facial image quality assessment method.

[0021] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. Through the artifact effect detection module of the present invention, the system can judge the existence of artifacts in real time, accurately identify the area where the artifacts are generated and its impact range, and avoid the misjudgment of image features caused by artifacts in the prior art. The regional division and artifact quality assessment module uses the brightness balance deviation index and the detail texture destruction index to quantitatively assess the specific impact of artifacts on each area, which not only improves the scientificity and accuracy of the assessment, but also provides a clear reference basis for the subsequent dynamic intervention mechanism, solving the problem of confusion in regional quality assessment in the prior art.

[0022] 2. The present invention can effectively improve the image quality in the artifact area by implementing targeted intervention measures on high-impact areas, medium-impact areas, and low-impact areas respectively. The quality of high-impact areas is significantly improved by strengthening brightness adjustment, texture enhancement, and edge smoothing, while moderate adjustment or monitoring is adopted for medium-impact and low-impact areas to avoid waste of resources. This hierarchical intervention method significantly improves processing efficiency while ensuring refined control of image quality, solving the problem of overall quality interference caused by unnecessary global adjustments in the prior art. The feedback mechanism during the intervention process further enhances the adaptability and accuracy of dynamic intervention through real-time monitoring and optimization, so that the system can maintain excellent performance under complex light source conditions.

[0023] 3. The present invention conducts comprehensive monitoring and analysis of various types of data obtained during the artifact adjustment and evaluation process, providing a solid foundation for the continuous improvement of artifact suppression methods and dynamic intervention strategies. By constructing a closed-loop optimization mechanism for the entire artifact processing process, the technical solution can achieve dynamic regulation of the entire process from artifact detection to quality restoration, significantly improving the overall effect of facial image quality assessment. Compared with traditional methods, this solution not only improves the reliability and accuracy of face recognition in dynamic light source environments, but also provides more intelligent and efficient technical support for actual scenarios such as advertising, thereby improving system performance while meeting the needs of commercial applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0025] Figure 1 The present invention is a flowchart of a facial image quality assessment method, system and computer-readable storage medium.

[0026] Figure 2 It is a module schematic diagram of the face image quality assessment method, system and computer-readable storage medium of the present invention. DETAILED DESCRIPTION

[0027] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0028] The present invention provides Figure 1 The face image quality assessment method shown in the figure specifically comprises the following steps: Under dynamic light source environment, the face image captured by the camera in real time is tested for artifact effects. By analyzing the uniformity of image brightness distribution, gradient direction anomalies, and texture feature changes, it is determined whether there is an artifact effect. Artifact effect refers to the phenomenon that in facial images, due to the strong reflection of dynamic light sources (such as neon lights or reflected light) on certain areas of the face (such as the forehead or nose bridge), and the influence of light propagation on adjacent areas (such as the eye area), the image has unnatural brightness transition, abnormal texture or loss of details. This effect will mask the true characteristics of the facial image and affect its reliability in quality assessment and recognition. In order to detect the artifact effect, the overall brightness distribution of the image, the edge characteristics of the local area and the integrity of the texture can be comprehensively analyzed to identify the interference characteristics of light reflection on different areas of the image and determine whether there is an artifact effect. This detection process can adapt to the changes of dynamic light sources in real time and provide support for subsequent processing.

[0029] By analyzing the uniformity of image brightness distribution, edge direction characteristics, and texture changes, the main interference of artifact effects on the image can be fully captured. Analysis of brightness distribution can identify overly bright or dark areas caused by light reflection, analysis of edge characteristics can find the damage of artifact effects to normal texture directions, and analysis of texture integrity can reveal the loss of details or patchy texture interference caused by artifacts. This multi-dimensional comprehensive analysis can effectively distinguish the difference between artifact effects and normal image features, ensuring the accuracy and comprehensiveness of the detection results.

[0030] Detecting artifact effects is a key step in this technical solution, and its purpose is to accurately identify quality problems caused by dynamic light reflections and provide basic support for subsequent area division and intervention. In a dynamic light source environment, artifact effects can cause abnormalities in local features and information loss, and existing technologies have difficulty effectively identifying such interference. If the artifact effects are not detected first, the subsequent area division and dynamic intervention mechanisms will lack pertinence, which may lead to distortion of quality assessment results. This solution provides an accurate basis for subsequent artifact classification, dynamic intervention and suppression mechanisms through real-time detection and precise judgment of artifact effects, solving the problem of insufficient processing of artifact effects in existing technologies.

[0031] When it is determined that there is an artifact effect, the face image captured by the camera in real time is evenly divided into several areas through a facial key point detection method; In the case of artifact effects, the facial image captured by the camera in real time is evenly divided into several areas through the facial key point detection method, in order to more accurately evaluate the impact of artifact effects on different parts of the face. The facial key point detection method can quickly identify the position and contour of key facial areas such as eyes, nose bridge, forehead, mouth and chin, and divide the facial image into several functional areas based on this, such as forehead, eye periphery, nose bridge, etc. At the same time, in order to capture the cross-region propagation phenomenon of artifact effects, boundary buffers can also be defined in adjacent areas (such as the junction of forehead and eyes) to achieve fine-grained division between regions. This division method can be implemented through software, using facial key point positioning algorithms (such as convolutional neural network models based on deep learning) to detect the feature points of the face, and using these key points as references, using regional grid division or polygon segmentation methods to complete the uniform division of the image. This regional division method can not only achieve localized management of artifacts, but also avoid quality assessment errors that may be caused by global analysis, ensuring that the artifact effects of each area can be evaluated independently. Combined with our technical solution, this division method can provide precise regional support for subsequent artifact quality assessment and dynamic intervention mechanism, solve the problem of assessment confusion caused by artifact effects affecting local features in existing technologies, and ultimately improve the accuracy and robustness of facial image quality assessment under dynamic light source environments.

[0032] Acquire artifact quality assessment information of each area of ​​the face image in real time, analyze it after acquisition, assess the impact of artifacts on the quality of each area, and divide each area into high-impact area, medium-impact area and low-impact area according to the assessment results; In this embodiment, artifact quality assessment information of each area of ​​the face image is obtained in real time, and analyzed after acquisition to assess the impact of artifacts on the quality of each area, and each area is divided into a high impact area, a medium impact area, and a low impact area according to the assessment results. Specifically, the following steps are included: Acquire artifact quality assessment information of each area of ​​the face image in real time and perform preprocessing after acquisition; Real-time acquisition of artifact quality assessment information can be achieved by dynamically analyzing the facial images captured by the camera. Specifically, the image data transmitted in real time by the camera is first segmented into regions, and the entire image is divided into multiple regions. Then, the brightness distribution characteristic information and texture structure characteristic information are extracted region by region through the image processing algorithm. The acquisition of brightness distribution characteristic information can be completed by calculating the brightness value statistics (such as mean and variance) of the pixels in each region, and the acquisition of texture structure characteristic information can be completed by applying a high-pass filter to extract the high-frequency detail components of the region, and combining the gradient calculation algorithm to analyze the texture direction characteristics and consistency of the region. The entire acquisition process is processed in frames to achieve real-time update of face image quality assessment information under dynamic light source environment.

[0033] Preprocessing is performed to ensure the accuracy and consistency of the evaluation information, thus laying the foundation for subsequent analysis. Preprocessing mainly includes three steps: noise filtering, brightness normalization, and boundary smoothing. Noise filtering removes noise interference in the image that may be introduced by the sensor or the external environment through median filtering or Gaussian filtering; brightness normalization linearly stretches or contrast stretches the brightness values ​​of each region to eliminate the overall brightness difference caused by changes in camera lighting; boundary smoothing corrects the mutation characteristics at the edge of the region through convolution operations, making cross-regional feature analysis more accurate. These preprocessing processes are implemented through software and can be gradually completed in a pipeline manner in a real-time image processing framework to ensure that the information extracted and analyzed later is more reliable.

[0034] Extracting brightness distribution characteristic information and texture structure characteristic information from artifact quality assessment information of each region of the preprocessed face image, and analyzing them after extraction to generate brightness balance deviation index and detail texture destruction index of each region respectively; Extracting the brightness distribution characteristic information and texture structure characteristic information of each region of the preprocessed face image can be achieved by performing pixel-by-pixel analysis on the regional image data through image processing algorithms. Specifically, the preprocessed face image is first divided into multiple functional regions (such as forehead, eyes, bridge of nose, etc.) according to the regional division rules, and the pixel data of each region is extracted. For the brightness distribution characteristic information, the brightness value of the pixels in each region is calculated pixel by pixel, and the average value, variance, and brightness difference between the boundary pixel and the neighboring boundary pixel are counted to reflect the brightness distribution state and boundary transition characteristics of the region; and for the texture structure characteristic information, the gradient vector of each pixel is calculated through the gradient calculation algorithm (such as Sobel operator), and the gradient amplitude, direction, and gradient change rate in the time series are further obtained, and the average value and distribution variance are counted in the region to reflect the significance and directional consistency of the texture. These operations are completed through image segmentation, pixel-by-pixel analysis and regional statistics, which can extract the comprehensive characteristics of brightness and texture in real time, and provide comprehensive characteristic data support for artifact impact assessment.

[0035] An artifact impact assessment model is constructed for the brightness balance deviation index and detail texture destruction index of each generated area, and the artifact impact coefficient of each area is generated. After the generation, the model is analyzed to evaluate the impact of the artifact on the quality of each area. According to the evaluation results, each area is divided into high-impact area, medium-impact area and low-impact area.

[0036] In this embodiment, the logic for obtaining the brightness balance deviation index and detail texture destruction index of each region is as follows: The brightness distribution characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average and variance of the brightness values ​​of all pixels in each area of ​​the face image at different times over a period of time, and the absolute average difference between the brightness values ​​of the edge pixels and the neighboring edge pixels of each area, and calibrated as , and , Indicates that within a period of time Moment face image The average brightness value of all pixels in the area Indicates that within a period of time Moment face image The variance of the brightness values ​​of all pixels in the region is Indicates that within a period of time Moment The absolute average difference between the brightness values ​​of the edge pixels in the region and the edge pixels in the neighborhood, , , and are all positive integers; The average brightness value of all pixels in each area of ​​the face image at different times within a period of time: First, in each frame of the image, the arithmetic mean (i.e., brightness mean) of the brightness values ​​of all pixels in each area is calculated, and then the brightness means at different times in the time series are averaged to obtain the brightness mean of the area in the entire time period, reflecting the dynamic characteristics of the overall brightness level of the area.

[0037] Variance of the brightness values ​​of all pixels in each area of ​​the face image at different times over a period of time: For the brightness data of each area in each frame of the image, the variance of the pixel brightness values ​​in the area is calculated to measure the intensity of the brightness change in the area, and then the variance data at different times is time-averaged to generate the temporal characteristics of the brightness variance of the area. This data reflects the degree of damage to the smoothness of the brightness distribution of the local area caused by the artifact.

[0038] The absolute average difference between the brightness values ​​of the edge pixels of each region and the neighboring edge pixels at different times over a period of time: Use the boundary detection algorithm to identify the boundary pixels of the current region and the neighboring region, calculate the absolute difference between the brightness values ​​of the boundary pixels of the current region and the boundary pixels of the neighboring region pixel by pixel, and average the differences of all boundary pixels to obtain the absolute average difference. This data dynamically reflects the cross-region brightness change characteristics caused by artifacts at the region boundary.

[0039] Calculate the brightness balance deviation index of each area. The specific calculation formula is as follows:

[0040] In the formula, For the Brightness balance deviation index of each area; Brightness Balance Deviation Index The calculation formula comprehensively evaluates the impact of artifacts on the brightness distribution of each area and the consistency of boundary brightness through multiple stages of calculation steps, thereby achieving comprehensive quantification of artifact interference. The formula takes the average value of the calculation results at multiple moments in a period of time, which can effectively eliminate the noise interference caused by instantaneous illumination fluctuations in a single frame image and ensure the stability of the evaluation results. Subsequently, the formula Partially by making the brightness difference and edge brightness deviation The product of Normalization evaluates the combined effect of artifacts on the brightness distribution within and at the boundaries of a region, while avoiding the amplification effect of extreme cases of excessively high or low brightness on the results. The term logarithmically amplifies the brightness deviation of the region boundary The contribution to the evaluation results further highlights the interference effect of artifacts in cross-regional propagation. Through the synergy of these calculation steps, the formula can fully capture the impact of artifacts on regional brightness balance and boundary consistency, providing scientific and accurate quantitative results.

[0041] No. Brightness balance deviation index of each area The size of directly reflects the degree of influence of the artifact on the quality of the area. The larger the index value, the more serious the interference of the artifact on the brightness distribution of the area. Specifically, The brightness fluctuation in the area is integrated ( ), regional edge brightness discontinuity ( ) and the overall brightness level ( ) dynamic characteristics. If A larger value indicates that the artifact causes uneven brightness changes in the area; if Higher values ​​indicate that artifacts at region boundaries cause cross-region light propagation interference; lower values ​​overall By combining these factors into an index value, It is able to quantify the extent to which artifacts disrupt brightness uniformity and boundary consistency. Areas with higher index values ​​are assessed to be more disturbed by artifact quality and may require stronger intervention measures.

[0042] The texture structure characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average change rate, amplitude and directional distribution variance of the pixel gradient in each area of ​​the face image at different times over a period of time, and calibrated as , and , Indicates that within a period of time Moment face image The average rate of change of pixel gradients in a region is Indicates that within a period of time Moment face image The magnitude of the pixel gradient in the region is Indicates that within a period of time Moment face image The directional distribution variance of pixel gradients in a region; The acquisition and interpretation of these three types of quantitative data are as follows: Gradient is a vector that describes the rate of change of pixel brightness in an image and is usually used to measure the brightness change in an image. The average rate of change of pixel gradient is obtained by calculating the brightness change of all pixels in the same area frame by frame, which is used to reflect the detail change characteristics caused by artifacts; specifically, over a period of time, artifacts will cause the gradient change in the area to become drastic or uneven, thereby affecting the detail characteristics. The amplitude of the pixel gradient represents the intensity of the pixel brightness change. The presence of artifacts usually reduces the gradient amplitude, making the image edge blurred and the texture insignificant. This data is obtained by analyzing the intensity of the brightness change in the area pixel by pixel and averaging the entire area. The directional distribution variance of the pixel gradient describes the directional consistency of the pixel brightness change in the area, and the texture directional destruction caused by artifacts is evaluated by analyzing the discrete degree of the gradient direction. Artifacts will cause disordered changes in the gradient direction, thereby significantly increasing the discreteness of the directional distribution. These three types of data comprehensively reflect the comprehensive impact of artifacts on the regional detail texture from the three perspectives of change rate, change intensity and directional consistency.

[0043] Calculate the detail texture destruction index of each area. The specific calculation formula is as follows:

[0044] In the formula, For the The detail texture destruction index of the area.

[0045] The formula comprehensively quantifies the degree of damage to the texture details of each area by artifacts through multiple calculation steps. It is well designed and targeted. First, the formula performs time averaging, which can smooth out the noise and sudden interference in a single frame image over a period of time to ensure the stability of the evaluation results. The first part of the formula The detail gradient change rate in the region is comprehensively considered ( ) and the gradient magnitude ( ), the product of the two quantifies the dynamic changes of regional details and edge significance. The presence of artifacts will reduce these values, while the gradient direction distribution variance in the denominator ( ) by amplifying the effect of disordered changes in orientation, making the artifact's contribution to the destruction of orientation consistency more significant. The second part of the formula The relationship between the disorder degree of directional distribution and the rate of detail change is further evaluated, and the square root operation is used to smooth the extreme values ​​that may be generated when the artifacts are serious to avoid distortion of the evaluation results. The design of the overall formula comprehensively reflects the degree of damage to the regional detail texture caused by the artifacts from the three dimensions of detail change, texture intensity and directional consistency. At the same time, the average calculation in the time dimension enhances the adaptability to dynamic light source scenes and the accuracy of the evaluation results.

[0046] No. Detail texture destruction index of the area The size of directly reflects the extent to which the artifact damages the detailed texture of the region, and is positively correlated with the impact of the artifact on the regional quality. Specifically, The larger the value, the more seriously the detail texture of the area is affected by the artifact, including abnormal detail change rate, reduced texture edge significance, and destroyed gradient direction consistency. A higher index value indicates that the texture direction disorder caused by the artifact is significant, and the texture details and edge characteristics in the area are severely damaged, which directly affects the image quality and reliability of feature extraction in the area. Through this index, the degree of interference of artifacts on the texture quality of each area can be quantified. Areas with higher indexes require stronger intervention measures to repair their quality problems.

[0047] In this embodiment, the brightness balance deviation index of each area is generated. and detail texture destruction index Construct an artifact impact assessment model and generate the artifact impact coefficient of each area through weighted summation , and the artifact influence coefficients of each region generated The threshold interval of the pre-set artifact influence coefficient Compare and evaluate the impact of artifacts on the quality of each area based on the comparison results, and divide each area into high-impact area, medium-impact area and low-impact area based on the evaluation results. The specific comparison analysis and division are as follows: like , if the artifact has a low impact on the quality of the area, then the area is classified as a low-impact area; This situation means that the artifact has a negligible impact on the quality of the area, and the brightness distribution and texture characteristics in the area remain basically stable, with no significant abnormalities or damage. Since the artifact interference is weak, the image quality of the area is close to the original state and does not require further repair or intervention. Such areas usually have no adverse effects on subsequent image analysis (such as face recognition or feature extraction) and can be directly used for related tasks, thereby improving the processing efficiency of the system and reducing the waste of computing resources.

[0048] like , if the artifact has a moderate impact on the quality of the area, then the area is classified as a medium-impact area; This indicates that the artifact has a significant impact on the quality of the area, but has not yet reached the level of serious damage. There may be certain brightness fluctuations or localized damage to texture characteristics in the area, such as blurred edges or slight disturbances in directional consistency, which may interfere with subsequent image analysis. For such areas, the system needs to take appropriate intervention measures based on the specific circumstances of the artifact impact, such as adjusting brightness or enhancing texture, to restore the image quality of the area as much as possible to ensure that it meets the requirements of subsequent analysis tasks.

[0049] like If the artifact has a high degree of impact on the quality of the area, the area is divided into a high-impact area.

[0050] This situation means that the artifact has a very serious impact on the quality of the area, and the brightness distribution and texture characteristics may have been significantly damaged, resulting in a significant decrease in the image quality of the area. For example, brightness overexposure or underexposure may occur, as well as texture loss or complete directional consistency disorder. If this area is not effectively repaired, it will significantly affect the accuracy of subsequent facial feature extraction or recognition. In order to reduce the interference of artifacts, the system must take strong intervention measures on high-impact areas, such as brightness correction, texture enhancement or artifact suppression algorithms, to maximize the restoration of image quality and ensure the normal progress of subsequent tasks.

[0051] The specific implementation of "building an artifact impact assessment model based on the brightness balance deviation index and detail texture destruction index of each region, and generating the artifact impact coefficient of each region by weighted summation" is as follows: First, the brightness balance deviation index of each region is and detail texture destruction index As input variables, weight coefficients are assigned based on their importance in artifact impact assessment to build a weighted assessment model. The weight coefficients can be set according to the different contributions of the two indicators to the artifact impact. For example, if the artifact in a scene has a more significant damage to the brightness balance, a higher weight is assigned to the brightness balance deviation index and a lower weight is assigned to the detail texture damage index. The sum of the weights is always 1 (e.g. and ,satisfy ). In the model, the formula Calculate the The artifact influence coefficient of each region ( ), and The model combines the brightness and texture characteristics into a unified quantitative indicator. By adjusting the weight coefficient, it can flexibly adapt to the main characteristics of artifact interference in different scenes, making the evaluation results more targeted and accurate, and providing a scientific basis for regional division and intervention measures.

[0052] The preset threshold interval of the artifact influence coefficient can be automatically determined by analyzing a large amount of sample data and combining statistical methods. The specific implementation method is as follows: First, collect a dataset of face images under a variety of typical dynamic light sources, covering areas with different artifact influence levels (such as low-impact, medium-impact, and high-impact areas). Then, the brightness balance deviation index and detail texture destruction index of each sample area are calculated through the image processing algorithm to further generate the artifact influence coefficient. The values ​​are labeled and classified according to their impact (e.g., manually labeled or using other validated quality assessment models to assist in classification), and the By calculating the value distribution of each category The mean and standard deviation of the values ​​determine the classification boundaries, such as the upper limit of the low-impact region ( ) can take low impact area The maximum value of the value, the lower limit of the high impact area ( ) can take high impact area The minimum value of the value can be further refined by setting quantiles (such as the 25th quantile and the 75th quantile). Finally, these thresholds are stored in the system as model parameters to form a dynamically adjustable threshold interval of the artifact impact coefficient, making the model more adaptable and accurate in classification under different environments or data distributions.

[0053] According to the division results of each area of ​​the face image, a dynamic intervention mechanism is constructed to take corresponding intervention measures for high-impact areas, medium-impact areas and low-impact areas respectively; In this embodiment, a dynamic intervention mechanism is constructed according to the division results of each area of ​​the face image, specifically: according to the division results of the high-impact area, the medium-impact area and the low-impact area, different intervention measure parameters are set respectively to form a dynamic intervention mechanism; the dynamic intervention mechanism is based on the artifact influence coefficient of each area, and automatically determines the adjustment method and amplitude of brightness adjustment, texture enhancement and edge smoothing through pre-set rules; The construction of dynamic intervention mechanism can be realized by combining the rule engine and image processing algorithm in the software system. The specific method is as follows: First, based on the calculation results of the artifact influence coefficient of each area of ​​the face image, the intervention strategies corresponding to the high-impact area, medium-impact area and low-impact area are preset through the rule engine. These strategies include the specific adjustment methods and amplitudes of brightness adjustment, texture enhancement and edge smoothing. The rule engine dynamically matches the intervention measures according to the range of the artifact influence coefficient, such as allocating enhanced parameters to the high-impact area, strengthening the brightness uniformity adjustment, enhancing the texture characteristics and smoothing the edge transition; allocating moderate optimization parameters to the medium-impact area, making moderate adjustments to the brightness and texture; and allocating monitoring parameters to the low-impact area, and only monitoring the quality status. In the specific implementation, the brightness normalization algorithm in the image processing module is called to adjust the brightness distribution consistency in the area, and the texture enhancement algorithm (such as local contrast enhancement or high-frequency detail amplification) is used to optimize the texture characteristics in the area, and the edge detection and smoothing algorithms are combined to eliminate the boundary discontinuity or mutation caused by artifacts. The purpose of this is to flexibly adjust intervention measures according to the degree of impact of artifacts on the quality of each area, ensure maximum resource utilization efficiency, and maintain regional stability in dynamic lighting environments while ensuring image quality, ultimately providing reliable image data support for face recognition and analysis tasks.

[0054] Corresponding intervention measures are taken for high-impact areas, medium-impact areas and low-impact areas respectively. Specifically, in high-impact areas, the artifact suppression enhancement parameters in the dynamic intervention mechanism are used to significantly adjust the brightness distribution, strengthen texture enhancement and optimize edge smoothing; in medium-impact areas, the artifact moderate adjustment parameters in the dynamic intervention mechanism are used to moderately optimize the brightness distribution and texture characteristics; in low-impact areas, the artifact stabilization parameters in the dynamic intervention mechanism are maintained without adjustment, and only the regional status is monitored to maintain stable quality.

[0055] Corresponding intervention measures are implemented for high-impact areas, medium-impact areas, and low-impact areas, respectively, which can be achieved through dynamic parameter control combined with image processing algorithms. Specifically, the software system first matches the intervention strategy of each area with the degree of artifact impact based on the artifact suppression parameters, medium adjustment parameters, or stability parameters assigned by the dynamic intervention mechanism. In high-impact areas, the brightness distribution is significantly adjusted through the brightness adjustment algorithm, such as eliminating overexposure or underexposure through histogram equalization or brightness stretching algorithms; high-frequency details are enhanced and texture characteristics are restored through texture enhancement algorithms (such as local contrast enhancement or Laplacian operators); at the same time, edge smoothing algorithms (such as Gaussian filtering or convolution operations) are combined to optimize boundary mutations caused by artifacts to ensure smooth transition of edge characteristics. In medium-impact areas, a moderate optimization strategy is adopted to make small adjustments to the brightness distribution and texture characteristics, such as moderate brightness mean normalization and contrast enhancement, to reduce the impact of artifacts but not excessive intervention, and ensure resource utilization efficiency. In low-impact areas, artifact stabilization parameters are used to maintain the current state, and only the brightness, texture, and edge characteristics of the area are dynamically monitored through a real-time monitoring algorithm to ensure the stability of the regional quality without increasing the computational burden. The purpose of implementing graded intervention in this way is to accurately adjust the intervention intensity according to the degree of artifact impact, maximize the image quality of high-impact areas, and avoid unnecessary operations on medium and low-impact areas, thereby balancing system performance and processing efficiency to ensure the balance and stability of the overall image quality.

[0056] In the process of dynamic intervention mechanism intervening in various regions, the intervention feedback information of each region is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations, and optimize the dynamic intervention mechanism according to the evaluation results; In this embodiment, during the process of the dynamic intervention mechanism intervening in each area, the intervention feedback information of each area is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each area can achieve the expected effect, and optimize the dynamic intervention mechanism according to the evaluation results, which specifically includes the following steps: In the process of dynamic intervention mechanism intervening in each area, the intervention feedback information of each area is obtained in real time and preprocessed after acquisition; Real-time acquisition of intervention feedback information of each area can be achieved by dynamically monitoring the real-time adjustment effect of the intervention mechanism on the image. Specifically, in the process of implementing dynamic intervention in each area, the system continuously captures the changes in brightness and texture characteristics of each area based on the image processing module. By analyzing the image data after intervention frame by frame, the pixel brightness value and gradient information of each area are extracted, including the brightness mean, brightness distribution characteristics, and key data such as texture gradient amplitude, gradient direction, and directional distribution variance of the area. These data are analyzed in real time by algorithms and attributed to each area to form comprehensive feedback information including brightness and texture adjustment effects. In addition, multi-frame feedback data is captured by time series processing to establish a continuous evaluation data set for dynamic intervention effects, ensuring that the feedback information can fully reflect the actual impact of the intervention on the image.

[0057] The purpose of preprocessing is to improve the accuracy and stability of feedback information and provide a reliable data basis for subsequent analysis. After the feedback information is obtained, a series of preprocessing operations need to be performed on it, mainly including denoising, normalization and outlier detection. Denoising can use Gaussian filtering or median filtering methods to eliminate random errors in image data that may be introduced by sensor noise or external environmental interference; normalization processing unifies the brightness values ​​and gradient amplitudes of each region to the same level through linear stretching or contrast adjustment, which is convenient for subsequent analysis and comparison; outlier detection uses statistical methods (such as mean and standard deviation range judgment) to eliminate possible erroneous data points to ensure the overall credibility of the feedback data. These preprocessing operations are gradually completed through the software pipeline to provide high-quality input data for the subsequent generation of brightness adjustment deviation index and texture recovery consistency index.

[0058] Extracting brightness adjustment feedback information and texture optimization feedback information from the preprocessed intervention feedback information of each region, and analyzing them after acquisition to generate a brightness adjustment deviation index and a texture restoration consistency index for each region respectively; Extracting the brightness adjustment feedback information and texture optimization feedback information from the intervention feedback information of each region after preprocessing can be achieved through a multidimensional data analysis method based on regional characteristics. Specifically, the preprocessed intervention feedback information is first divided into regions, and the brightness and texture-related characteristic data in each region are extracted respectively. For the brightness adjustment feedback information, by analyzing the statistical characteristics of the pixel brightness values ​​in the region, data including the brightness mean and brightness variance are extracted to measure the adjustment effect of the overall brightness of the region and the consistency of the brightness distribution. For the texture optimization feedback information, the gradient amplitude and directional distribution characteristics of each region are calculated by the image gradient algorithm, and the texture intensity and directional consistency characteristics of the region after intervention are extracted, including key data such as the mean value of the gradient amplitude and the variance of the gradient direction. These data are calculated pixel by pixel and summarized by region to form the brightness adjustment feedback information and texture optimization feedback information of each region, and are uniformly managed through a standardized storage structure, providing detailed regional characteristic data support for subsequent indicator calculation and intervention effect evaluation.

[0059] An intervention effect evaluation model is constructed for the brightness adjustment deviation index and texture restoration consistency index generated for each area, and the intervention coefficient of each area is generated. The generated intervention coefficient of each area is compared with the pre-set intervention coefficient threshold of each area. According to the comparison results, it is evaluated whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and the dynamic intervention mechanism is optimized according to the evaluation results.

[0060] The pre-set intervention coefficient thresholds for each area can be achieved through statistical analysis of historical data. The specific method is to collect facial image samples before and after the intervention in multiple scenarios, calculate the brightness adjustment deviation index and texture restoration consistency index of each area, and generate the intervention coefficient. The intervention coefficients are classified and labeled according to the intervention effect of the samples, and the mean, median and standard deviation of the intervention coefficient distribution of various intervention effects are calculated through statistical analysis. The distribution median or mean of the efficient intervention sample is used as the ideal range, and the threshold range is set in combination with the quantile or ± standard deviation to ensure rationality and adaptability. At the same time, the threshold setting can be dynamically adjusted according to the needs of different scenarios or optimized through machine learning.

[0061] In this embodiment, the logic for obtaining the brightness adjustment deviation index and the texture restoration consistency index of each area is as follows: The brightness adjustment feedback information in the intervention feedback information of each area after preprocessing is extracted, specifically including the average value and variance of the pixel brightness values ​​of each area at different times within a period of time after the intervention, and the preset average value of the pixel brightness values ​​of each area after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average value of the pixel brightness in the region Indicates that within a period of time after the intervention Moment The variance of pixel brightness values ​​in the region, The pre-specified intervention The average value of the pixel brightness in the region , , and are all positive integers; Within a period of time after the intervention, the average value, variance and pre-set target brightness value of the pixel brightness values ​​of each area at different times can be dynamically obtained through image processing algorithms and system preset parameters. First, the system captures the image data after the intervention in real time through the camera, and divides the face image into multiple independent functional areas (such as forehead, eyes, nose bridge, etc.) according to the regional division rules, and extracts the pixel brightness value data for each area frame by frame. For the average value of the pixel brightness value, it is obtained by calculating the arithmetic average of all the pixel brightness values ​​in the area in each frame; for the variance of the pixel brightness value, it is completed by calculating the degree of deviation of each pixel brightness value in the area from the brightness mean of the area and calculating the variance, which is used to reflect the uniformity of the brightness distribution. The above two types of data are collected at multiple times within a period of time after the intervention, and averaged according to the time series to dynamically generate core data reflecting the effect of regional brightness distribution adjustment. As for the target brightness value, it is a quantitative data pre-set according to the scene requirements or ideal quality goals when the intervention mechanism is set, which serves as an important reference for comparing and evaluating the intervention effect. This acquisition method not only ensures the dynamic real-time nature of the data, but also combines the guidance of the target value, providing a scientific basis for the quantitative evaluation of the brightness adjustment effect.

[0062] Calculate the brightness adjustment deviation index of each area. The specific calculation formula is as follows:

[0063] In the formula, For the Brightness adjustment deviation index for each region; The design of calculating the brightness adjustment deviation index in the above way is mainly to comprehensively and dynamically evaluate the effect of the dynamic intervention mechanism on the adjustment of regional brightness distribution. By obtaining the brightness data at multiple times within a period of time after the intervention and taking the average value, the fluctuation and noise interference that may exist in the instantaneous data can be effectively eliminated, ensuring that the calculation results are more stable and reliable. The two parts of the formula - the brightness mean deviation and the brightness distribution variance - quantify the overall deviation and distribution uniformity of the brightness adjustment respectively. This design can capture the global adjustment effect and reflect the detailed characteristics of local brightness inconsistency. At the same time, normalizing the brightness variance and combining it with the brightness mean deviation can more intuitively show whether the consistency of the brightness distribution has been improved after the intervention. Especially in the dynamic light source scene, this calculation method can more comprehensively reflect the intervention effect and provide an accurate and quantifiable basis for subsequent optimization.

[0064] No. Brightness adjustment deviation index for each region The size of directly reflects whether the brightness adjustment effect of the dynamic intervention mechanism in this area meets expectations. The larger the index, the higher the degree of deviation of the brightness distribution after intervention from the target value, and the worse the adjustment effect; the smaller the index, the closer the brightness adjustment is to the target value, and the more ideal the effect. Specifically, The first part quantifies the deviation between the mean brightness after intervention and the target value. If the deviation is large, it means that the overall brightness adjustment has not reached the ideal level. The second part reflects the uniformity of brightness distribution after intervention through brightness variance. If the distribution is uneven, it indicates that the local brightness adjustment is still insufficient. By comparing the index value with the pre-set evaluation threshold range, it can be clarified whether the effect of the dynamic intervention mechanism in this area exceeds expectations, meets expectations, or fails to meet expectations, providing a scientific basis for further optimizing the intervention strategy.

[0065] The texture optimization feedback information in the intervention feedback information of each region after preprocessing is extracted, including the average amplitude of the pixel gradient of each region at different times within a period of time after the intervention, the directional distribution variance, and the pre-set average amplitude of the pixel gradient of each region after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average magnitude of the pixel gradient in the region, Indicates that within a period of time after the intervention Moment The directional distribution variance of the pixel gradient in the region, The pre-specified intervention The average amplitude of pixel gradients in a region; The real-time image processing algorithm can be used to obtain the average amplitude of the pixel gradients in each region at different times, the directional distribution variance, and the pre-set average amplitude of the pixel gradients in each region after the intervention. First, the gradient vector of each pixel is calculated for each pixel point in each region using the gradient calculation algorithm (such as the Sobel operator or the Scharr operator) through the image captured by the camera in real time after the intervention, and the gradient amplitude and gradient direction are extracted from it. For the average amplitude of the gradient, the gradient amplitude of all pixels in the region is averaged to reflect the texture intensity level of the region; for the directional distribution variance, the variance of the gradient direction of all pixels in the region is calculated to quantify the consistency of the texture direction of the region after the intervention; and the pre-set average amplitude of the gradient is based on the intervention goal. The ideal gradient amplitude is set under the condition of artifact elimination, which can usually be determined by analyzing the artifact-free image data set or expert experience. These data comprehensively describe the texture characteristics of each region after the intervention from three dimensions: brightness change, texture saliency, and directional consistency, providing an accurate quantitative basis for evaluating the intervention effect.

[0066] Calculate the texture restoration consistency index of each area. The specific calculation formula is as follows:

[0067] In the formula, For the Texture restoration consistency index for each region.

[0068] The calculation result of texture restoration consistency index is to take the average value of data at multiple moments in a period of time as the purpose of comprehensively and stably reflecting the overall restoration effect of dynamic intervention on regional texture characteristics. The texture characteristics of each region after intervention may gradually stabilize over time or show short-term fluctuations in different frames. By averaging the gradient amplitude, directional distribution variance and other data at multiple moments, the interference of instantaneous fluctuations on the results can be effectively smoothed, and the evaluation error caused by accidental or noise data at a single moment can be avoided. In addition, this time dimension processing can more truly reflect the overall trend and cumulative impact of the intervention effect, so that the calculation results can not only evaluate the immediate state of texture restoration after intervention, but also reflect the long-term effect, providing a more reliable and scientific basis for further optimizing the intervention mechanism.

[0069] No. Texture restoration consistency index of the region The size of directly reflects the degree of deviation between the intervention effect of the dynamic intervention mechanism in this area and the expected goal. Specifically, The smaller the value, the closer the texture characteristics of the area after intervention (such as gradient amplitude and directional consistency) are to the preset target value, indicating that the effect of the dynamic intervention mechanism in this area is closer to the expectation; conversely, The larger the value, the less texture strength or more disordered directional distribution of the area after intervention, indicating that the intervention effect has not fully restored the quality of the area and there is still much room for improvement. The size of can be used to quantify the degree of success of texture restoration after intervention, and based on this, it can be judged whether the dynamic intervention mechanism has achieved the set quality goals in each area, providing a clear direction for optimizing the intervention strategy.

[0070] In this embodiment, the brightness of each area is adjusted to the deviation index. and texture recovery consistency index Construct an intervention effect evaluation model and generate intervention coefficients for each region through weighted summation , and the intervention coefficients of each region generated The intervention coefficient thresholds for each area are set in advance Compare and evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations based on the comparison results, and optimize the dynamic intervention mechanism based on the evaluation results. The specific comparison and analysis are as follows: like , the intervention effect of the dynamic intervention mechanism in this area can achieve the expected effect, and there is no need to optimize the dynamic intervention mechanism; This situation means that the dynamic intervention mechanism has achieved the expected intervention effect in this area, and the brightness distribution and texture characteristics of the area have been successfully restored to the ideal state or close to the target level. The brightness adjustment deviation index and texture recovery consistency index are both within an acceptable range, indicating that the impact of the artifacts after the intervention has been effectively eliminated or significantly weakened, and will not cause negative interference to the image quality of the area. In this case, there is no need to further adjust the dynamic intervention mechanism, and the current parameter settings can be maintained, which not only saves computing resources, but also ensures the efficiency of the system operation, and provides high-quality data support for subsequent image analysis.

[0071] like The intervention effect of the dynamic intervention mechanism in this area cannot meet the expectations, and the dynamic intervention mechanism needs to be optimized, including: increasing the brightness adjustment intensity of the area and reducing the brightness distribution deviation; increasing the parameter weight of texture enhancement, enhancing the contrast of details and high-frequency texture characteristics; adjusting the edge smoothing algorithm to reduce the boundary mutation problem caused by artifact interference; combining the real-time intervention feedback data of the area, dynamically adjusting the intervention parameters and rules, and optimizing the execution process of the intervention mechanism.

[0072] This situation indicates that the dynamic intervention mechanism has not achieved the expected intervention effect in this area, and there are still large deviations in the brightness and texture characteristics of the area. For example, insufficient brightness adjustment or failure to restore the consistency of texture direction may result in significant artifact interference. Such results may interfere with subsequent image processing links (such as feature extraction or face recognition) and reduce the accuracy and reliability of the system. At this point, it is necessary to optimize the parameters and rules of the dynamic intervention mechanism, such as strengthening brightness adjustment, increasing texture enhancement intensity, or adjusting edge smoothing algorithms, to further improve the image quality of the area and ensure that the overall effect meets expectations.

[0073] These optimization measures can be achieved by combining image processing algorithms and dynamic parameter control, as follows: First, through the brightness adjustment module in the software system, combined with the real-time regional brightness distribution data, the weights and parameters of the brightness equalization algorithm are dynamically adjusted, such as using adaptive histogram equalization (CLAHE) or brightness normalization methods to significantly enhance the contrast of regional brightness and reduce brightness deviation, thereby improving the brightness consistency after intervention. Secondly, texture enhancement optimization can be achieved through local contrast enhancement or algorithms based on high-frequency detail extraction (such as Laplace enhancement or wavelet transform), which can enhance the texture characteristics in the region by improving the significance of high-frequency details, and adjust the weights of enhancement parameters to make texture recovery more significant and close to the target effect. Edge smoothing optimization can smooth the boundary mutations caused by artifact interference through methods such as Gaussian filtering or bilateral filtering, further improving the transition effect and visual quality of the edge area. In addition, combined with real-time intervention feedback data (such as brightness distribution deviation value, texture characteristic index, etc.), the key parameters of the intervention mechanism and the execution order of the algorithm are dynamically adjusted through the rule engine, such as determining the priority of brightness adjustment and texture enhancement or adjusting the enhancement intensity based on feedback data. This approach not only optimizes the intervention strategy according to the regional state, but also dynamically optimizes the intervention effect through continuous adaptive adjustment. Through these methods, the intervention effect can be significantly improved, ensuring that the impact of artifacts is minimized while keeping the brightness and texture characteristics of the area close to the target state.

[0074] “Building an intervention effect evaluation model for the brightness adjustment deviation index and texture restoration consistency index of each region, and generating the intervention coefficient of each region by weighted summation” can be achieved in the following ways: First, the brightness adjustment deviation index of each region and texture recovery consistency index As the input variable of the evaluation model, it reflects the comprehensive effect of the dynamic intervention mechanism on brightness distribution and texture restoration. and , construct the weighted sum formula ,in and The value of is determined by the importance of the two indicators in a specific scene. For example, in a scene with higher requirements for brightness distribution, The weight of , giving a greater contribution to brightness adjustment; in the scene that emphasizes texture detail recovery, it can be increased The weights of the intervention effect are calculated to make the contribution of texture restoration to the evaluation of intervention effect more significant. These weight coefficients can be set by experiments or scene requirements to ensure the flexibility and applicability of the model. Finally, the weighted intervention coefficient of each area is calculated. , comprehensively evaluate whether the intervention effect in the area has met expectations, and provide a quantitative basis for further optimization of the dynamic intervention mechanism.

[0075] Real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process are carried out to continuously improve artifact suppression methods and dynamic intervention strategies and optimize the facial image quality assessment process.

[0076] This goal can be achieved by combining real-time data acquisition, analysis models and optimization algorithms. Specifically, first, through the data acquisition module in the software system, artifact quality assessment information (such as brightness distribution characteristics, texture characteristics, etc.), intervention feedback information (such as brightness adjustment results and texture enhancement effects, etc.), intervention records (including intervention parameter settings and rule execution) and image quality restoration effects (such as changes in artifact influence coefficients in each region) are obtained in real time. These data are dynamically stored in the system's database and passed as input to the comprehensive analysis module. Then, the data is associated with the analysis model, such as using the time series analysis model to predict the trend of the artifact adjustment effect, or mining the correlation between the intervention parameters and the artifact assessment results based on the rule engine to identify potential optimization points in the current dynamic intervention strategy. At the same time, the historical intervention records are trained in combination with machine learning algorithms (such as regression analysis or decision trees) to form a data-driven optimization recommendation model. Finally, the execution rules of the artifact suppression method and the parameter configuration of the intervention strategy are dynamically adjusted through the optimization module, so that the system can enhance the accuracy of artifact adjustment and the adaptability of dynamic intervention in continuous learning and improvement. The purpose of doing this is to continuously optimize the overall process of facial image quality assessment through real-time monitoring and comprehensive analysis of the entire artifact processing data, so that the system can more efficiently adapt to artifact interference under complex lighting conditions and ensure the accuracy and stability of subsequent face recognition and feature extraction.

[0077] like Figure 2 The facial image quality assessment system shown includes an artifact effect detection module, a region division and calibration module, an artifact quality assessment module, a dynamic intervention execution module, an intervention effect optimization module, and a comprehensive monitoring improvement module; The artifact effect detection module performs artifact effect detection on the face image captured by the camera in real time under a dynamic light source environment. It determines whether there is an artifact effect by analyzing the uniformity of the image's brightness distribution, abnormal gradient direction, and changes in texture features. The region division and calibration module evenly divides the face image captured by the camera in real time into several regions through the facial key point detection method when it is determined that there is an artifact effect; The artifact quality assessment module obtains the artifact quality assessment information of each area of ​​the face image in real time, analyzes it after acquisition, evaluates the impact of the artifact on the quality of each area, and divides each area into high-impact area, medium-impact area and low-impact area according to the assessment results; The dynamic intervention execution module builds a dynamic intervention mechanism based on the division results of each area of ​​the face image, and takes corresponding intervention measures for the high-impact area, medium-impact area and low-impact area respectively; The intervention effect optimization module obtains the intervention feedback information of each area in real time during the process of the dynamic intervention mechanism intervening in each area, analyzes it after obtaining, evaluates whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and optimizes the dynamic intervention mechanism according to the evaluation results; The comprehensive monitoring and improvement module conducts real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process, continuously improves the artifact suppression method and dynamic intervention strategy, and optimizes the facial image quality assessment process.

[0078] A computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a facial image quality assessment method.

[0079] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0080] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0081] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0082] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0083] In the several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0084] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0085] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0086] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for assessing the quality of a face image, characterized in that: The specific steps include: Under dynamic light source environment, the face image captured by the camera in real time is tested for artifact effects. By analyzing the uniformity of image brightness distribution, gradient direction anomalies, and texture feature changes, it is determined whether there is an artifact effect. When it is determined that there is an artifact effect, the face image captured by the camera in real time is evenly divided into several areas through a facial key point detection method; Acquire artifact quality assessment information of each area of ​​the face image in real time, analyze it after acquisition, assess the impact of artifacts on the quality of each area, and divide each area into high-impact area, medium-impact area and low-impact area according to the assessment results; According to the division results of each area of ​​the face image, a dynamic intervention mechanism is constructed to take corresponding intervention measures for high-impact areas, medium-impact areas and low-impact areas respectively; In the process of dynamic intervention mechanism intervening in various regions, the intervention feedback information of each region is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations, and optimize the dynamic intervention mechanism according to the evaluation results; Real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process are carried out to continuously improve artifact suppression methods and dynamic intervention strategies and optimize the facial image quality assessment process.

2. The facial image quality assessment method according to claim 1, characterized in that: The artifact quality assessment information of each area of ​​the face image is obtained in real time, and analyzed after acquisition to assess the impact of the artifact on the quality of each area, and the areas are divided into high-impact areas, medium-impact areas and low-impact areas according to the assessment results. The specific steps include: Acquire artifact quality assessment information of each area of ​​the face image in real time and perform preprocessing after acquisition; Extracting brightness distribution characteristic information and texture structure characteristic information from artifact quality assessment information of each region of the preprocessed face image, and analyzing them after extraction to generate brightness balance deviation index and detail texture destruction index of each region respectively; An artifact impact assessment model is constructed for the brightness balance deviation index and detail texture destruction index of each generated area, and the artifact impact coefficient of each area is generated. After the generation, the model is analyzed to evaluate the impact of the artifact on the quality of each area. According to the evaluation results, each area is divided into high-impact area, medium-impact area and low-impact area.

3. The facial image quality assessment method according to claim 2, characterized in that: The logic for obtaining the brightness balance deviation index and detail texture destruction index of each area is as follows: The brightness distribution characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average and variance of the brightness values ​​of all pixels in each area of ​​the face image at different times over a period of time, and the absolute average difference between the brightness values ​​of the edge pixels and the neighboring edge pixels of each area, and calibrated as , and , Indicates that within a period of time Moment face image The average brightness value of all pixels in the area Indicates that within a period of time Moment face image The variance of the brightness values ​​of all pixels in the region is Indicates that within a period of time Moment The absolute average difference between the brightness values ​​of the edge pixels in the region and the edge pixels in the neighborhood, , , and All are positive integers; Calculate the brightness balance deviation index of each area. The specific calculation formula is as follows: In the formula, For the Brightness balance deviation index of each area; The texture structure characteristic information in the artifact quality assessment information of each area of ​​the preprocessed face image is extracted, including the average change rate, amplitude and directional distribution variance of the pixel gradient in each area of ​​the face image at different times over a period of time, and calibrated as , and , Indicates that within a period of time Moment face image The average rate of change of pixel gradients in a region is Indicates that within a period of time Moment face image The magnitude of the pixel gradient in the region, Indicates that within a period of time Moment face image The directional distribution variance of pixel gradients in a region; Calculate the detail texture destruction index of each area. The specific calculation formula is as follows: In the formula, For the The detail texture destruction index of the area.

4. The facial image quality assessment method according to claim 3, characterized in that: Brightness balance deviation index for each generated area and detail texture destruction index Construct an artifact impact assessment model and generate the artifact impact coefficient of each area through weighted summation , and the artifact influence coefficients of each region generated The threshold interval of the pre-set artifact influence coefficient Compare and evaluate the impact of artifacts on the quality of each area based on the comparison results, and divide each area into high-impact area, medium-impact area and low-impact area based on the evaluation results. The specific comparison analysis and division are as follows: like , if the artifact has a low impact on the quality of the area, then the area is classified as a low-impact area; like , if the artifact has a moderate impact on the quality of the area, then the area is classified as a medium-impact area; like If the artifact has a high degree of impact on the quality of the area, the area is classified as a high-impact area.

5. The method for assessing facial image quality according to claim 4, characterized in that: According to the division results of each area of ​​the face image, a dynamic intervention mechanism is constructed, specifically: according to the division results of the high-impact area, the medium-impact area and the low-impact area, different intervention measure parameters are set respectively to form a dynamic intervention mechanism; the dynamic intervention mechanism is based on the artifact influence coefficient of each area and automatically determines the adjustment method and amplitude of brightness adjustment, texture enhancement and edge smoothing through pre-set rules; Corresponding intervention measures are taken for high-impact areas, medium-impact areas and low-impact areas respectively. Specifically, in high-impact areas, the artifact suppression enhancement parameters in the dynamic intervention mechanism are used to significantly adjust the brightness distribution, strengthen texture enhancement and optimize edge smoothing; in medium-impact areas, the artifact moderate adjustment parameters in the dynamic intervention mechanism are used to moderately optimize the brightness distribution and texture characteristics; in low-impact areas, the artifact stabilization parameters in the dynamic intervention mechanism are maintained without adjustment, and only the regional status is monitored to maintain stable quality.

6. The method for assessing facial image quality according to claim 5, characterized in that: In the process of dynamic intervention mechanism intervening in various regions, the intervention feedback information of each region is obtained in real time, and analyzed after acquisition to evaluate whether the intervention effect of the dynamic intervention mechanism in each region can achieve the expected effect, and optimize the dynamic intervention mechanism according to the evaluation results, which specifically includes the following steps: In the process of dynamic intervention mechanism intervening in each area, the intervention feedback information of each area is obtained in real time and preprocessed after acquisition; Extracting brightness adjustment feedback information and texture optimization feedback information from the preprocessed intervention feedback information of each region, and analyzing them after acquisition to generate a brightness adjustment deviation index and a texture restoration consistency index for each region respectively; An intervention effect evaluation model is constructed for the brightness adjustment deviation index and texture restoration consistency index generated for each area, and the intervention coefficient of each area is generated. The generated intervention coefficient of each area is compared with the pre-set intervention coefficient threshold of each area. According to the comparison results, it is evaluated whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and the dynamic intervention mechanism is optimized according to the evaluation results.

7. The method for assessing facial image quality according to claim 6, characterized in that: The logic for obtaining the brightness adjustment deviation index and texture restoration consistency index of each area is as follows: The brightness adjustment feedback information in the intervention feedback information of each area after preprocessing is extracted, specifically including the average value and variance of the pixel brightness values ​​of each area at different times within a period of time after the intervention, and the preset average value of the pixel brightness values ​​of each area after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average value of the pixel brightness in the region Indicates that within a period of time after the intervention Moment The variance of the pixel brightness values ​​in the region, The pre-specified intervention The average value of the pixel brightness in the region , , and All are positive integers; Calculate the brightness adjustment deviation index of each area. The specific calculation formula is as follows: In the formula, For the Brightness adjustment deviation index for each region; The texture optimization feedback information in the intervention feedback information of each region after preprocessing is extracted, including the average amplitude of the pixel gradient of each region at different times within a period of time after the intervention, the directional distribution variance, and the pre-set average amplitude of the pixel gradient of each region after the intervention, and they are calibrated as , and , Indicates that within a period of time after the intervention Moment The average magnitude of the pixel gradient in the region, Indicates that within a period of time after the intervention Moment The directional distribution variance of the pixel gradient in the region, The pre-specified intervention The average amplitude of the pixel gradient in the region; Calculate the texture restoration consistency index of each area. The specific calculation formula is as follows: In the formula, For the Texture restoration consistency index for each region.

8. The method for assessing facial image quality according to claim 7, characterized in that: Adjust the brightness of each area generated by the deviation index and texture recovery consistency index Construct an intervention effect evaluation model and generate intervention coefficients for each region through weighted summation , and the intervention coefficients of each region generated The intervention coefficient thresholds for each area are set in advance Compare and evaluate whether the intervention effect of the dynamic intervention mechanism in each region can meet expectations based on the comparison results, and optimize the dynamic intervention mechanism based on the evaluation results. The specific comparison and analysis are as follows: like , the intervention effect of the dynamic intervention mechanism in this area can achieve the expected effect, and there is no need to optimize the dynamic intervention mechanism; like The intervention effect of the dynamic intervention mechanism in this area cannot meet the expectations, and the dynamic intervention mechanism needs to be optimized, including: increasing the brightness adjustment intensity of the area and reducing the brightness distribution deviation; increasing the parameter weight of texture enhancement, strengthening the contrast of details and high-frequency texture characteristics; combining the real-time intervention feedback data of the area, dynamically adjusting the intervention parameters and rules, and optimizing the execution process of the intervention mechanism.

9. A facial image quality assessment system, used to implement the facial image quality assessment method according to any one of claims 1 to 8, characterized in that: It includes artifact effect detection module, area division and calibration module, artifact quality assessment module, dynamic intervention execution module, intervention effect optimization module and comprehensive monitoring improvement module; The artifact effect detection module performs artifact effect detection on the face image captured by the camera in real time under a dynamic light source environment. It determines whether there is an artifact effect by analyzing the uniformity of the image's brightness distribution, abnormal gradient direction, and changes in texture features. The region division and calibration module evenly divides the face image captured by the camera in real time into several regions through the facial key point detection method when it is determined that there is an artifact effect; The artifact quality assessment module obtains the artifact quality assessment information of each area of ​​the face image in real time, analyzes it after acquisition, evaluates the impact of the artifact on the quality of each area, and divides each area into high-impact area, medium-impact area and low-impact area according to the assessment results; The dynamic intervention execution module builds a dynamic intervention mechanism based on the division results of each area of ​​the face image, and takes corresponding intervention measures for the high-impact area, medium-impact area and low-impact area respectively; The intervention effect optimization module obtains the intervention feedback information of each area in real time during the process of the dynamic intervention mechanism intervening in each area, analyzes it after obtaining, evaluates whether the intervention effect of the dynamic intervention mechanism in each area can meet the expectations, and optimizes the dynamic intervention mechanism according to the evaluation results; The comprehensive monitoring and improvement module conducts real-time monitoring and comprehensive analysis of the artifact quality assessment information, intervention feedback information, intervention records and image quality restoration effects obtained during the artifact adjustment and evaluation process, continuously improves the artifact suppression method and dynamic intervention strategy, and optimizes the facial image quality assessment process.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the facial image quality assessment method according to any one of claims 1 to 8.

Images