Handwriting regularity monitoring method and system based on image enhancement

Abstract

The application discloses a handwriting standard monitoring method and system based on image enhancement, relates to the technical field of image processing and intelligent evaluation, and comprises the following steps: obtaining a handwriting character image through adaptive light compensation of multiple light sources; after image preprocessing such as geometric correction and adaptive binarization, a convolutional neural network is used to extract character skeleton topological structure features for structural standard analysis; meanwhile, a pen tip motion trajectory is extracted based on an optical flow method and time sequence standard analysis is performed; after the structural and time sequence bimodal features are fused, a scoring model constructed based on a style transfer network is input to calculate a standard score; and a hierarchical error correction report is automatically generated according to the score grade and deviation type. The technical problems of incomplete stroke extraction, insufficient structural analysis precision, lack of dynamic writing feature modeling and lack of hierarchical error correction feedback under the complex illumination and shielding conditions in the prior art are solved.

Description

Technical Field

[0001] This application relates to the fields of image processing and intelligent evaluation technology, and in particular to a method and system for monitoring handwriting standardization based on image enhancement. Background Technology

[0002] Handwriting accuracy monitoring is one of the core technologies in the field of intelligent education, with wide applications in early literacy instruction, calligraphy practice, and special education. However, existing handwriting accuracy monitoring technologies still have many limitations, making it difficult to achieve accurate and comprehensive automated assessment and guidance.

[0003] First, at the image acquisition and preprocessing level, existing technologies struggle to handle complex real-world writing scenarios. Uneven lighting, paper reflections, and shadows in the writing environment severely degrade the quality of acquired images, leading to incomplete, broken, or fused stroke extraction, directly impacting the accuracy of feature analysis. Conventional image binarization and denoising methods have poor adaptability and lack robust solutions for complex optical interference.

[0004] Secondly, existing methods are often rather simplistic in their normative analysis. Most techniques focus only on the static structural norms of writing, such as evaluating stroke shape and structure through template matching or simple contour comparison, completely ignoring the dynamic temporal norms of writing, namely the correctness of stroke order and the rhythm of pen strokes. However, stroke order and pen strokes are crucial to the formation of writing habits, and errors in these areas, if not corrected in time, will affect writing speed and the aesthetics of the characters. Furthermore, traditional structural analysis methods are mostly based on the characteristics of hand-designed structures, lacking the ability to perceive subtle structural deviations such as the uniformity of stroke width and the consistency of spacing between substructure units, and lacking a deep utilization of the topological features of the skeleton.

[0005] Furthermore, in terms of evaluation models, existing scoring mechanisms typically rely on simple rules or shallow machine learning models, making it difficult to quantify the ineffable "standardization" or "consistency" inherent in the writing results. This leads to a gap between the scoring results and the professional judgment of human teachers, resulting in a lack of persuasiveness.

[0006] Finally, in the feedback and guidance phase, most existing systems can only provide simple right / wrong judgments or single-pattern demonstrations, failing to generate tiered, targeted error correction solutions based on the user's error type and severity. Learners cannot obtain personalized reinforcement training content such as targeted copying templates or dynamic writing order demonstrations, thus significantly reducing the teaching effectiveness.

[0007] Therefore, there is an urgent need in this field for a comprehensive solution that can overcome the above-mentioned defects, achieve high-quality image acquisition in complex scenarios, integrate structural and temporal normative analysis, perform accurate and human-centered intelligent scoring, and ultimately provide targeted hierarchical error correction guidance. Summary of the Invention

[0008] The purpose of this application is to provide a method and system for monitoring handwriting conformity based on image enhancement. This aims to address the technical problems in existing technologies, such as incomplete stroke extraction under complex lighting and occlusion conditions, insufficient accuracy of structural analysis, lack of dynamic writing feature modeling, and lack of hierarchical error correction feedback.

[0009] In view of the above technical problems, this application provides a method and system for monitoring handwriting standardization based on image enhancement.

[0010] A first aspect of this application provides a handwriting conformity monitoring method based on image enhancement, the method comprising:

[0011] The original image of the handwritten character is acquired by an image acquisition device equipped with multi-source adaptive lighting function. Geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation are performed on the original image of the handwritten character in sequence to obtain a binary image with continuous strokes.

[0012] The binarized image is input into a convolutional neural network to extract the skeleton topological features of the characters. Stroke width uniformity parameters and substructure unit spacing consistency parameters are extracted from the skeleton topological features. The skeleton topological features are compared with standard character templates to generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include the structural deviation type and the position information of the structural deviation type in the handwritten character image.

[0013] The pen tip movement trajectory during the writing process is extracted using optical flow. Based on the trajectory point sequence marked with continuous timestamps, the changes in motion direction and instantaneous velocity between adjacent trajectory points are analyzed. Based on the geometric characteristics of the writing motion, a dynamic temporal feature analysis step is used to assign a higher evaluation weight to high curvature trajectory segments than to straight trajectory segments. The consistency of writing sequence norms and pen rhythm is analyzed through temporal matching to obtain temporal norms indexes and temporal deviation analysis results. The temporal deviation analysis results include the type of temporal deviation and the position information of the type of temporal deviation in the writing sequence.

[0014] The structural standardization index and the temporal standardization index are weighted and fused according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result, and the temporal deviation analysis result, are input into a scoring model trained by machine learning to calculate the standardization score.

[0015] Based on the standardization score, the structural deviation analysis results, and the temporal deviation analysis results, the error correction intensity and guidance content are automatically determined, and a tiered error correction report is generated and output. The tiered error correction report includes targeted copying templates and writing order demonstrations.

[0016] A second aspect of this application provides a handwriting conformity monitoring system based on image enhancement, the system comprising:

[0017] The image acquisition module is used to acquire the original image of the handwritten character through an image acquisition device equipped with multi-source adaptive lighting function, and to sequentially perform geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation on the original image of the handwritten character to obtain a binary image with continuous strokes.

[0018] An image enhancement processing module is used to input the binarized image into a convolutional neural network, extract the skeleton topological features of the characters, extract stroke width uniformity parameters and substructure unit spacing consistency parameters from the skeleton topological features, compare the skeleton topological features with standard character templates, and generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include structural deviation types and the location information of the structural deviation types in the handwritten character image.

[0019] The feature extraction and analysis module is used to extract the pen tip movement trajectory during the writing process using optical flow. Based on the trajectory point sequence marked with continuous timestamps, it analyzes the changes in movement direction and instantaneous speed between adjacent trajectory points. Based on the geometric features of the writing motion, it assigns a higher evaluation weight to high-curvature trajectory segments than to straight trajectory segments through a dynamic temporal feature analysis step. It also analyzes the writing sequence standardization and pen rhythm consistency through temporal matching to obtain temporal standardization index and temporal deviation analysis results. The temporal deviation analysis results include the type of temporal deviation and the position information of the type of temporal deviation in the writing sequence.

[0020] The comprehensive scoring module is used to weight and fuse the structural standardization index and the temporal standardization index according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result, and the temporal deviation analysis result, are input into a scoring model trained based on machine learning to calculate the standardization score.

[0021] The error correction report generation module is used to automatically determine the error correction intensity and guidance content based on the standardization score level, the structural deviation analysis results and the temporal deviation analysis results, and generate and output a hierarchical error correction report. The hierarchical error correction report includes a targeted copying template and a demonstration of the writing order.

[0022] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0023] By employing multi-source adaptive illumination and image preprocessing, the technology effectively overcomes interference from complex environments, significantly improving stroke extraction quality. It innovatively integrates structural and temporal dual-modal analysis to achieve a comprehensive evaluation of both the static shape of handwriting and the dynamic writing process. A dynamic weight allocation mechanism based on high-curvature trajectory segments enhances the accuracy of deviation analysis for key writing areas. A scoring model based on a style transfer network quantifies writing standardization features into an objective measure of style consistency, making the scoring results more convincing. A hierarchical error correction report generation mechanism enables personalized and adaptive feedback guidance based on error type and severity. Finally, through complete hardware and software system integration, the technology achieves large-scale application capabilities, significantly improving the quality and efficiency of handwriting instruction. This technology solves the technical problems of incomplete stroke extraction under complex lighting and occlusion conditions, insufficient accuracy of structural analysis, lack of dynamic writing feature modeling, and lack of hierarchical error correction feedback in existing technologies.

[0024] The above description is merely an overview of the technical solution of this application. In order to more clearly explain the technical means of this application, and to enable its implementation in accordance with the contents of the specification, and to make the above and other objectives, features and advantages of this application more apparent and understandable, specific embodiments of this application are described below. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0026] Figure 1 A flowchart illustrating the image-enhanced handwriting conformity monitoring method provided in this application embodiment;

[0027] Figure 2 This is a schematic diagram of the structure of the image-enhanced handwriting standardization monitoring system provided in an embodiment of this application.

[0028] Figure labeling: Image acquisition module 10, Image enhancement processing module 20, Feature extraction and analysis module 30, Comprehensive scoring module 40, Error correction report generation module 50. Detailed Implementation

[0029] This application provides a handwriting standardization monitoring method and system based on image enhancement, which solves the technical problems in the prior art such as incomplete stroke extraction under complex lighting and occlusion conditions, insufficient accuracy of structural analysis, lack of dynamic writing feature modeling, and lack of hierarchical error correction feedback.

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0031] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0032] Example 1, as Figure 1 As shown, this application provides a handwriting standardization monitoring method based on image enhancement, wherein the method includes:

[0033] The original image of the handwritten character is acquired by an image acquisition device equipped with multi-source adaptive lighting function. Geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation are performed on the original image of the handwritten character in sequence to obtain a binary image with continuous strokes.

[0034] Furthermore, the morphological opening operation uses a 5×5 rectangular structural element;

[0035] Furthermore, the image acquisition device includes:

[0036] An image sensor with 24-bit analog-to-digital conversion capability is used to improve the color depth and brightness resolution of the original image of handwritten characters;

[0037] The light source module with adaptive supplemental lighting function is used to automatically adjust the supplemental lighting brightness according to the ambient light intensity in order to suppress paper reflection and shadow occlusion.

[0038] Specifically, the original image of the handwritten character is acquired by an image acquisition device equipped with multi-source adaptive supplementary lighting. The image acquisition device uses a 12-megapixel image sensor (IMX477) with 24-bit analog-to-digital conversion, a 35mm lens focal length, and an aperture of F1.8. The light source module consists of four independently controllable LED supplementary lights arranged around the lens. The ambient illuminance is monitored in real time by a TSL2591 ambient light sensor. When the ambient illuminance is below 300 lux, the supplementary lighting is automatically turned on. 300 lux is selected as the supplementary lighting trigger threshold because this illuminance is the optimal lighting threshold for handwriting scenarios (below 300 lux, stroke shadows are likely to appear; above 300 lux, paper reflections are likely to occur). A 32×32 pixel local image block is used to balance binarization accuracy and processing speed (too small a block is easily affected by noise, while too large a block is likely to lose stroke details). The supplementary lighting intensity is linearly adjusted within the range of 100-500 lux based on the ambient illuminance. The image sensor is connected to the main processor via a MIPI CSI-2 interface with a data transmission rate of 2.5 Gbps. After acquiring the original image of the handwritten character, the following processing is performed on the 1280×720 pixel RGB image: First, geometric distortion correction is performed using the camera intrinsic parameter matrix obtained by Zhang Zhengyou calibration method to eliminate radial and tangential distortion; then, adaptive binarization processing based on local image patch brightness statistics is performed, with a local image patch size of 32×32 pixels, and the optimal threshold for each local region is determined using the maximum inter-class variance method, with a threshold calculation window sliding step of 16 pixels; finally, morphological opening processing is performed, which uses a 5×5 rectangular structuring element and is implemented using the cv2.morphologyEx function of OpenCV. First, an erosion operation is performed with the 5×5 rectangular structuring element, and then a dilation operation is performed with the same 5×5 rectangular structuring element. The processing time is 15 ms per frame, and finally, a binary image with continuous strokes is obtained. This embodiment runs on an Intel Core i5-8300H processor, and the entire preprocessing process takes less than 50ms, ensuring real-time processing capabilities.

[0039] The binarized image is input into a convolutional neural network to extract the skeleton topological features of the characters. Stroke width uniformity parameters and substructure unit spacing consistency parameters are extracted from the skeleton topological features. The skeleton topological features are compared with standard character templates to generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include the structural deviation type and the position information of the structural deviation type in the handwritten character image.

[0040] Furthermore, the standard character template includes standard skeleton topology information and standard writing order timing information for each character. The standard writing order timing information includes the writing order, direction, and standard speed curve of each stroke.

[0041] Furthermore, the binarized image is input into a convolutional neural network to extract the skeleton topological features of the characters. From these features, stroke width uniformity parameters and sub-structural unit spacing consistency parameters are extracted. The skeleton topological features are then compared with standard character templates to generate structural regularity indices and structural deviation analysis results, including:

[0042] By using a convolutional neural network based on the ResNet-50 architecture, the spatial coordinates of key points of character strokes in the binarized image are regressed, and the spatial coordinates of the key points are used to construct the skeleton topological features, wherein the key points include stroke endpoints, intersections, and high curvature points.

[0043] Based on the spatial coordinates of the key points, adjacent key points are connected by a linear interpolation algorithm to generate continuous stroke center lines.

[0044] Based on the character structure specification, key points that are spatially adjacent and topologically related are clustered into sub-structural units to obtain sub-structural unit segmentation information;

[0045] Based on the stroke center line, the variance of the width of the connected pixel regions on both sides of its normal direction is calculated to obtain the stroke width uniformity parameter.

[0046] Based on the sub-structural unit segmentation information, the Euclidean distance between the center points of the minimum bounding rectangle of the sub-structural unit is calculated and compared with the standard spacing of the corresponding sub-structural unit in the standard character template. The relative deviation value is calculated to obtain the sub-structural unit spacing consistency parameter.

[0047] Calculate the Hausdorff distance between the skeleton topological features and the corresponding keypoint set in the standard character template to generate a structural deviation matrix;

[0048] The stroke width uniformity parameter, the substructure unit spacing consistency parameter, and the structural deviation matrix are weighted and summed to generate a structural standardization index; based on the deviation values ​​and position information of each key point in the structural deviation matrix, structural deviation analysis results are generated.

[0049] Specifically, the binarized image obtained by preprocessing is input into a convolutional neural network based on the ResNet-50 architecture to extract the skeleton topological structure features of characters. Specifically, the ResNet-50 network removes the fully connected layers of the original model, and two parallel fully connected layers with 1024 dimensions are connected after the last convolutional layer, which are respectively used to regress the x coordinates and y coordinates of key points; the key points include stroke endpoints, intersection points, and high-curvature points, where the high-curvature points are defined as local extreme points where the curvature of the center line of the stroke skeleton exceeds 0.15, and the curvature is calculated by the three-point method (taking the central sampling point and one adjacent sampling point on each side, fitting an arc and then calculating the curvature); the network input is a 256×256 pixel binarized image, and the output is a sequence composed of the (x, y) coordinates of N key points. The spatial coordinates of the key points constitute the skeleton topological structure features. This feature is normalized before being compared with the standard character template, so that the width and height of the character are uniformly mapped into the 256×256 pixel coordinate system, and the coordinate accuracy error is less than 2 pixels, where N takes 10 - 30 according to the character complexity (for example, N = 6 for the character '人' and N = 28 for the character '繁'). Based on the spatial coordinates of the key points, adjacent key points are connected by a linear interpolation algorithm, specifically using the cubic spline interpolation method, and 10 interpolation points are used to connect adjacent key points to generate a continuous stroke center line. Based on the character structure specification, key points that are adjacent in space (Euclidean distance less than 15 pixels) and topologically related (judging connectivity through the Delaunay triangulation algorithm) are clustered into sub-structure units to obtain sub-structure unit segmentation information. Based on the stroke center line, the variance of the width of the connected pixel regions on both sides of its normal direction is calculated. The specific method is as follows: a sampling point is taken every 5 pixels along the stroke center line, and the distance between the intersection points of the normal direction of this point and the stroke boundary is calculated, and a total of M width values are obtained. M takes 20 - 50 according to the stroke length (for example, M = 20 for a short horizontal stroke and M = 50 for a long vertical stroke), and the variance of these M width values is calculated to obtain the stroke width uniformity parameter. Based on the sub-structure unit segmentation information, the Euclidean distance between the center points of the minimum bounding rectangles of each sub-structure unit is calculated and compared with the standard spacing of the corresponding sub-structure unit in the standard character template, and the relative deviation value (the calculation formula is: (measured spacing - standard spacing) / standard spacing × 100%) is calculated to obtain the sub-structure unit spacing consistency parameter. The Hausdorff distance between the skeleton topological structure feature and the corresponding key point set in the standard character template is calculated to generate a structure deviation matrix. The Hausdorff distance calculation uses the bidirectional maximum distance method, and the matrix dimension is N×N, where N is the number of key points.The stroke width uniformity parameter, the sub-structural unit spacing consistency parameter, and the structural deviation matrix are weighted and summed (weight coefficients are 0.3, 0.3, and 0.4, respectively; these weights are optimized using 1000 validation set samples covering different writing levels and different characters, with the goal of achieving the minimum mean square error between machine scoring and human scoring by three senior primary school Chinese teachers), generating structural standardization indicators. Based on the deviation values ​​and position information of each key point in the structural deviation matrix, structural deviation analysis results are generated. The structural deviation types include key point position offset (deviation value > 3 pixels is considered offset) and sub-structural unit spacing anomaly (relative deviation > 15% is considered anomaly). The deviation positions are marked in the form of bounding boxes in the handwritten character image, with the bounding box color gradually changing from yellow (slight 10%-20%) to red (severe deviation > 30%) according to the severity of the deviation. The standard character template is built based on the GB2312-80 character set and contains standard skeleton topology information and standard writing order timing information for 6763 Chinese characters. The standard writing order timing information includes the writing order, direction, and standard speed curve of each stroke. The standard speed curve is recorded at a sampling frequency of 100Hz, and the speed unit is pixels per second. This embodiment runs on an NVIDIA GTX1080Ti graphics card, and the total time for structural feature extraction and comparison is less than 80ms.

[0050] The pen tip movement trajectory during the writing process is extracted using optical flow. Based on the trajectory point sequence marked with continuous timestamps, the changes in motion direction and instantaneous velocity between adjacent trajectory points are analyzed. Based on the geometric characteristics of the writing motion, a dynamic temporal feature analysis step is used to assign a higher evaluation weight to high curvature trajectory segments than to straight trajectory segments. The consistency of writing sequence norms and pen rhythm is analyzed through temporal matching to obtain temporal norms indexes and temporal deviation analysis results. The temporal deviation analysis results include the type of temporal deviation and the position information of the type of temporal deviation in the writing sequence.

[0051] Furthermore, the dynamic temporal feature analysis step assigns a higher evaluation weight to high-curvature trajectory segments than to straight trajectory segments, and through temporal matching analysis of the consistency between writing order standardization and pen stroke rhythm, obtains temporal standardization indicators and temporal deviation analysis results, including:

[0052] Based on the continuously extracted pen tip motion trajectory using the optical flow method, a sequence of trajectory points marked by continuous timestamps is obtained;

[0053] The ratio of coordinate displacement to time interval between adjacent trajectory points is continuously calculated to obtain the change in motion direction and instantaneous velocity, and the direction change sequence and velocity change sequence are updated in real time.

[0054] Based on the statistical characteristics of the writing direction change, the high curvature region is detected and marked on the input direction change sequence. The high curvature region is defined as a continuous trajectory segment where the direction change exceeds a preset direction threshold.

[0055] Based on the state of the detected high curvature region, an evaluation weight coefficient is dynamically assigned to the trajectory segment. The weight coefficient includes a first evaluation weight coefficient assigned to the trajectory segment in the high curvature region and a second evaluation weight coefficient assigned to the trajectory segment in the non-high curvature region. The first evaluation weight coefficient is greater than the second evaluation weight coefficient.

[0056] The trajectory point sequence is weighted based on the first evaluation weight coefficient and the second evaluation weight coefficient to obtain a weighted trajectory point sequence.

[0057] Based on the weighted trajectory point sequence, the minimum cumulative path distance between the pen tip movement trajectory and the standard writing order time sequence information in the standard character template is calculated by the dynamic time warping algorithm, and the minimum cumulative path distance is mapped to the trajectory order deviation.

[0058] By comparing the velocity of the weighted trajectory point sequence with the distribution difference of the standard velocity curve in the standard writing sequence time information, a velocity consistency index is calculated.

[0059] Based on the trajectory sequence deviation, the speed consistency index, the position information of the abnormal trajectory segment corresponding to the trajectory sequence deviation in the pen tip movement trajectory, and the position information of the abnormal speed segment corresponding to the speed consistency index in the pen tip movement trajectory, a temporal regularity index and a temporal deviation analysis result are generated.

[0060] Specifically, the pen tip movement trajectory during the writing process is continuously extracted using the Farneback optical flow method at a sampling frequency of 60Hz, obtaining a sequence of trajectory points marked by continuous timestamps. Each trajectory point contains three-dimensional coordinate information (x, y, t). The coordinates of the trajectory points are normalized to linearly scale all coordinates to the reference coordinate system of the standard template; simultaneously, the timestamps are normalized to scale the total duration of the entire writing process to the total duration of the standard template, eliminating interference from differences in writing speed and size on timing matching. The ratio of coordinate displacement to time interval between adjacent trajectory points is continuously calculated to obtain the change in motion direction (in degrees / second) and the instantaneous velocity change (in pixels / second). The direction change sequence and velocity change sequence are updated in real time using a circular buffer with a capacity of 300 sampling points. Based on the statistical characteristics of changes in writing direction, high curvature regions are detected and labeled in the input sequence of changes in direction. Specifically, a sliding window variance analysis method is used, with a window size of 15 sampling points and a step size of 5 sampling points. The high curvature region is defined as a continuous trajectory segment in which the variance of the change in direction exceeds a preset direction threshold (set to 45 degrees / second, which is determined by statistical data of high curvature trajectory segments from 100 different writers (including children and adults), covering 95% of normal writing turning scenarios and avoiding misjudgment or omission). The length of the continuous trajectory segment must be greater than 3 sampling points. Based on the detected state of the high-curvature region, evaluation weight coefficients are dynamically assigned to the trajectory segments. These weight coefficients include a first evaluation weight coefficient (set to 1.5, verified through comparative experiments: compared to the unweighted scheme, it can improve the accuracy of deviation recognition in high-curvature regions by 22%, aligning with the actual scenario of "errors easily occurring at turning points" in writing) assigned to the trajectory segments in non-high-curvature regions, and a second evaluation weight coefficient (set to 0.8, forming a reasonable gradient with the first weight, highlighting key areas while avoiding excessive interference from non-key areas in the overall evaluation). The weight assignment response time is less than 10ms. The trajectory point sequence is then weighted based on the first and second evaluation weight coefficients to obtain a weighted trajectory point sequence. Based on the weighted trajectory point sequence, the minimum cumulative path distance between the pen tip movement trajectory and the standard writing order temporal information in the standard character template is calculated using a dynamic time warping algorithm. Dynamic time warping uses Euclidean distance as a local cost metric, and the curved path constraint uses Sakoe-Chiba bandwidth, with the bandwidth parameter set to 15% of the trajectory length (this bandwidth ratio was optimized through 500 sets of trajectory alignment experiments at different writing speeds, which can reduce computational complexity while ensuring alignment accuracy). Subsequently, the trajectory order deviation is generated: the minimum cumulative path distance obtained above is mapped to the [0,1] interval using a sigmoid function. Specifically, a logistic function is used, and its formula is: ,in To minimize the cumulative path distance, For function offset parameters, This is the function scaling parameter. and The value of is determined based on the statistical analysis of the minimum cumulative path distance of all samples in the training dataset: Set to the sample mean of this distance distribution (calculated to be approximately 5). The standard deviation of the distance distribution is set to the sample standard deviation (calculated to be approximately 2). This parameter setting aligns the center of the logistic function with the center of the data distribution, and its steepness matches the dispersion of the data, thus ensuring that the distance values ​​are optimally mapped to a trajectory order deviation between 0 and 1 (0 indicates no deviation, 1 indicates severe deviation). Then, a velocity consistency index is generated: the distribution difference between the velocity of the weighted trajectory point sequence and the standard velocity curve in the standard written time sequence information is compared. Using the dynamically time-warped aligned velocity sequence, the Pearson correlation coefficient between the two is calculated. Subsequently, through the linear transformation formula... correlation coefficient Mapping to the [0,1] interval yields a speed consistency index, where 0 indicates completely inconsistent speeds and 1 indicates completely consistent speeds. Based on the trajectory sequence deviation, the speed consistency index, the position information of the abnormal trajectory segment corresponding to the trajectory sequence deviation in the pen tip movement trajectory (determined through dynamic time warping path backtracking), and the position information of the abnormal speed segment corresponding to the speed consistency index in the pen tip movement trajectory (determined through sliding window correlation coefficient analysis, with a window size of 20 sampling points, and a correlation coefficient < 0.3 indicating an abnormal speed segment), a weighted summation method is used (weight coefficients are 0.6 and 0.4, respectively, which are determined through cross-validation of 100 sets of manually labeled temporal deviation samples, so that the correlation coefficient between the temporal standardization index and the manual score reaches 0.91) to generate a temporal standardization index; at the same time, the types of temporal deviations (including stroke order reversal, stroke order omission, and pen movement rhythm that is too fast / too slow, where stroke order reversal refers to the actual writing order being completely opposite to the standard order, stroke order omission refers to the absence of a stroke in the standard writing, and abnormal pen movement rhythm refers to the speed fluctuation amplitude exceeding 50% of the standard speed curve) and their specific positions in the writing sequence (marked by start and end timestamps) are recorded to generate temporal deviation analysis results.

[0061] The structural standardization index and the temporal standardization index are weighted and fused according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result, and the temporal deviation analysis result, are input into a scoring model trained by machine learning to calculate the standardization score.

[0062] Furthermore, the scoring model trained based on machine learning calculates a prescriptive score, including:

[0063] The comprehensive normative value, stroke width uniformity parameter, and substructure unit spacing consistency parameter are normalized to generate a feature vector with uniform dimensions.

[0064] The positional information in the structural deviation analysis results and the temporal deviation analysis results is encoded into a spatial weight matrix, and the spatial weight matrix corresponds one-to-one with the spatial position of the strokes and sub-structural units in the handwritten character image;

[0065] The scoring model trained by the machine learning is a prescriptive scoring model based on a style transfer network. The feature vector is forward propagated through the prescriptive scoring model to output a global stroke thickness uniformity score and a global substructure unit spacing coordination score.

[0066] Based on the spatial weight matrix, the global stroke thickness uniformity score is decomposed and mapped to the corresponding position of each stroke. The local uniformity score of each stroke is obtained by combining the spatial weights corresponding to each position.

[0067] Based on the spatial weight matrix, the global substructure unit spacing coordination score is decomposed and mapped to the corresponding position of each substructure unit. The local coordination score of each substructure unit is obtained by combining the spatial weights corresponding to each position.

[0068] The local uniformity scores of all strokes and the local coordination scores of all substructural units are weighted and summed, and the summation results are mapped to a scale from level 1 to level 10 to generate the final normalization score.

[0069] Specifically, the structural standardization index and the temporal standardization index are weighted and fused according to preset weights (structural weight 0.6, temporal weight 0.4, which were determined by statistically averaging the scoring weights of 500 sets of handwritten samples by 10 calligraphy education experts, with the structural index having a consensus of 85% in terms of weight in regular script, clerical script, and other fonts) to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis results, and the temporal deviation analysis results, are input into a scoring model trained based on machine learning to calculate the standardization score. Specifically, the comprehensive normalization value, stroke width uniformity parameter, and sub-structural unit spacing consistency parameter are first normalized. The min-max normalization method is used to linearly transform each feature value to the [0,1] interval, generating a 64-dimensional feature vector with unified dimensions (dimension composition: 1-dimensional comprehensive normalization value + 31-dimensional stroke width uniformity parameter (1 dimension per stroke) + 32-dimensional sub-structural unit spacing consistency parameter (1 dimension per sub-structural unit), with the total dimension dynamically adapted to the character complexity). The positional information in the structural deviation analysis results and the temporal deviation analysis results is encoded into a spatial weight matrix. The specific encoding rule is as follows: for each detected deviation position, a 15×15 pixel Gaussian kernel function (standard deviation σ=3) is established centered on it for weight assignment. The weight value is 1.0 at the center and decays to 0.1 towards the periphery (0.1 is an empirical threshold to ensure that the weight of the normal area outside the deviation area by 5 pixels is negligible). The spatial weight matrix corresponds one-to-one with the spatial positions of strokes and sub-structural units in the handwritten character image. The scoring model trained by the machine learning is a prescriptive scoring model based on a style transfer network. This model uses the VGG-19 network as its basic architecture, removes its last fully connected layer, and retains the first 16 convolutional layers (corresponding to conv1_1 to conv5_4 layers, retaining the multi-scale texture feature extraction capability required for style transfer) as feature extractors. Two parallel fully connected layers (dimensions of 128 and 64, respectively) are connected to regress the global stroke thickness uniformity score and the global substructure unit spacing coordination score, respectively. The feature vector is forward propagated through the prescriptive scoring model, using the ReLU activation function, with a dropout rate set to 0.2 (which has been verified to reduce the risk of overfitting and improve the test set accuracy by 4%). The global stroke thickness uniformity score and the global substructure unit spacing coordination score are output, and the initial value range of both scores is [0,1].Based on the spatial weight matrix, the global stroke thickness uniformity score is decomposed and mapped to the corresponding position of each stroke. Specifically, the decomposition method is as follows: a base score is allocated based on the proportion of the stroke centerline length to the total stroke length of the character (e.g., if a stroke's length accounts for 20%, then 20% of the global score is allocated as the base value). This base score is then multiplied by the spatial weight value corresponding to each position, and a weighted calculation is performed based on the spatial weights corresponding to each position to obtain the local uniformity score for each stroke. Based on the spatial weight matrix, the global substructure unit spacing coordination score is decomposed and mapped to the corresponding position of each substructure unit. The decomposition method is as follows: a base score is evenly allocated to each substructure unit (experiments have verified that in 90% of commonly used characters, the contribution of substructure units to coordination is similar, with an average allocation error of <5%). This base score is then multiplied by the average spatial weight of the corresponding region of each unit, and a weighted calculation is performed based on the spatial weights corresponding to each position to obtain the local coordination score for each substructure unit. The local uniformity scores of all strokes and the local coordination scores of all sub-structural units are weighted and summed (stroke score weight 0.7, sub-structural unit score weight 0.3, the weights are optimized on the validation set using a grid search method to minimize the mean squared error between the scores and expert annotations). The summation result is then mapped to a scale from level 1 to level 10 using a linear transformation y=9x+1 to generate the final standardization score. This scoring model is trained on a dataset containing 100,000 handwritten samples and their expert standardization scores (samples cover Chinese characters, numbers, and English letters, including different writing levels of children and adults; expert scores are independently annotated by more than 3 senior calligraphy teachers and the average is taken). The mean squared error loss function and the Adam optimizer (learning rate 0.001, batch size 32) are used, with a training cycle of 100 epochs. The consistency correlation coefficient between the model and expert scores on the test set reaches 0.93. This embodiment runs on an NVIDIA GTX 1080 Ti graphics card, and the total time for the scoring calculation process is less than 25ms.

[0070] Based on the level of the normative score, the results of the structural deviation analysis, and the results of the temporal deviation analysis, the error correction intensity and guidance content are automatically determined, and a tiered error correction report is generated and output. The tiered error correction report includes targeted copying templates and writing order demonstrations.

[0071] Furthermore, based on the level of the normative scoring, the results of the structural deviation analysis, and the results of the temporal deviation analysis, the automatic determination of the correction intensity and guidance content, and the generation and output of a hierarchical correction report, including:

[0072] The normative score, the structural deviation type from the structural deviation analysis results, and the temporal deviation type from the temporal deviation analysis results are all input into a predefined mapping rule.

[0073] The mapping rule defines the error correction intensity level corresponding to the combination of normative scoring level and deviation type based on the scoring interval division, so as to output the error correction intensity level of high intensity, medium intensity or low intensity.

[0074] Based on the deviation position coordinates in the structural deviation analysis results, standard images of corresponding strokes or sub-structural units are extracted from the standard character template to generate the first copying template.

[0075] Based on the abnormal trajectory segment position information in the temporal deviation analysis results, the standard writing order temporal information of the corresponding stroke is called from the standard character template to generate the first writing order demonstration animation;

[0076] Based on the error correction intensity level, the first copy template and the first writing order demonstration animation are differentiated to obtain an enhanced copy template and writing order demonstration animation;

[0077] The differentiated processing includes, when the error correction intensity level is high, superimposing visual guidance on the pen stroke direction and force on the first copying template and reducing the playback speed of the first writing order demonstration animation; when the error correction intensity level is medium, superimposing visual guidance on the pen stroke direction on the first copying template and appropriately reducing the playback speed of the first writing order demonstration animation; and when the error correction intensity level is low, directly outputting the first copying template and the first writing order demonstration animation as the basic content.

[0078] The enhanced copying template and writing order demonstration animation are combined with the standardization score, the structural deviation analysis results, and the temporal deviation analysis results to generate the hierarchical error correction report.

[0079] Specifically, the normative score, the structural deviation type in the structural deviation analysis results, and the temporal deviation type in the temporal deviation analysis results are jointly input into a predefined mapping rule. The mapping rule is stored in the form of a lookup table and is specifically defined as follows: when the normative score is 1-3 and there is component misalignment (relative deviation > 15%) or stroke order reversal, it is mapped to a high-intensity error correction level; when the normative score is 4-6 and there is a stroke width uniformity deviation exceeding the threshold (variance > 0.12 mm²) or a pen speed fluctuation exceeding the threshold (fluctuation amplitude > 50%), it is mapped to a medium-intensity error correction level; when the normative score is 7-10 and there is only slight stroke edge irregularity (edge ​​pixel deviation ≤ 2), it is mapped to a low-intensity error correction level; when the normative score is 7-10 and there is only slight stroke edge irregularity, it is mapped to a low-intensity error correction level. Based on the deviation location coordinates (e.g., a rectangular area from (x1, y1) to (x2, y2)) in the structural deviation analysis results, standard images of corresponding strokes or sub-structural units are extracted from the standard character template to generate the first tracing template. Specifically, the OpenCV cv2.rectangle() function is used to delineate the deviation area, and then the ROI extraction function is performed using cv2.imcrop() to extract a standard image of a 64×64 pixel area centered on the deviation location coordinates (the 64×64 pixel size is designed based on "covering the deviation area + 10 pixels of context" to ensure that the surrounding structure can be referenced during tracing and to avoid structural fragmentation caused by local extraction). Based on the abnormal trajectory segment location information (e.g., the trajectory point sequence corresponding to timestamps t1 to t2) in the temporal deviation analysis results, the standard writing order temporal information of the corresponding strokes is called from the standard character template to generate the first writing order demonstration animation. The animation frame rate is 24fps, the resolution is 320×240 pixels, and the animation base speed is set to 100 pixels / second (corresponding to normal writing speed). Speed ​​control is achieved by adjusting the frame interval. Based on the error correction intensity level, the first tracing template and the first writing order demonstration animation are differentiated: when the intensity level is high, a red arrow (8 pixels long, 2 pixels wide) and a blue gradient (from RGB(0,0,255) to RGB(0,128,255)) are superimposed on the first tracing template to indicate the pen stroke direction, and the playback speed of the first writing order demonstration animation is reduced to 50% of the base speed; when the intensity level is medium, a black arrow (5 pixels long, 1 pixel wide) is superimposed on the first tracing template to indicate the pen stroke direction, and the playback speed is adjusted to 75% of the base speed; when the intensity level is low, the first tracing template and the first writing order demonstration animation are directly output.The enhanced copying template and writing order demonstration animation are combined with the standardized scoring, structural deviation analysis results (including structural deviation types and deviation location coordinates), and temporal deviation analysis results (including temporal deviation types and timestamps corresponding to abnormal trajectory segment location information) to generate the hierarchical error correction report. Specifically, the copying template is stored in PNG format, and the writing order demonstration animation is stored in MP4 format, and they are associated using a unified file naming rule. The naming format is defined as "Character ID_User ID_Score X_Intensity Level_Template.png" and "Character ID_User ID_Score X_Intensity Level_Animation.mp4". The standardized scoring and deviation analysis results are recorded in JSON format. The JSON structure includes the score value, a detailed description of each deviation type, the filenames of the corresponding template and animation, and the server storage path to ensure that all components can be accurately matched and called by the system. When this embodiment is executed on an Intel Core i5-8300H processor, the total report generation time is less than 100ms, ensuring the real-time interactive performance of the system.

[0080] In summary, the embodiments of this application have at least the following technical effects:

[0081] The original image of handwritten characters is acquired using an image acquisition device equipped with multi-source adaptive illumination. Geometric distortion correction, adaptive binarization based on local image block brightness statistics, and morphological opening are sequentially performed on the original handwritten character image to obtain a binary image with continuous strokes. This binary image is then input into a convolutional neural network to extract the skeleton topological features of the characters. From these features, stroke width uniformity parameters and sub-structural unit spacing consistency parameters are extracted. The skeleton topological features are compared with standard character templates to generate structural standardization indicators and structural deviation analysis results. These results include the type of structural deviation and its location information within the handwritten character image. The pen tip movement trajectory during the writing process is extracted using optical flow. Based on a sequence of trajectory points marked with continuous timestamps, the changes in motion direction and instantaneous velocity between adjacent trajectory points are analyzed. Finally, based on the geometric features of the writing motion, dynamic... The temporal feature analysis step assigns a higher evaluation weight to high-curvature trajectory segments than to straight trajectory segments. Through temporal matching analysis of writing order standardization and pen stroke rhythm consistency, temporal standardization indicators and temporal deviation analysis results are obtained. The temporal deviation analysis results include the type of temporal deviation and its position information in the writing sequence. The structural standardization indicators and temporal standardization indicators are weighted and fused according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, sub-structural unit spacing consistency parameter, structural deviation analysis results, and temporal deviation analysis results, are input into a machine learning-trained scoring model to calculate a standardization score. Based on the level of the standardization score, the structural deviation analysis results, and the temporal deviation analysis results, the error correction intensity and guidance content are automatically determined, generating and outputting a layered error correction report. The layered error correction report includes targeted copying templates and writing order demonstrations.

[0082] Example 2, based on the same inventive concept as the image enhancement-based handwriting standardization monitoring method in the previous examples, such as... Figure 2 As shown, this application provides a handwriting standardization monitoring system based on image enhancement. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0083] Image acquisition module 10 is used to acquire the original image of handwritten characters through an image acquisition device equipped with multi-source adaptive lighting function, and to sequentially perform geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation on the original image of handwritten characters to obtain a binary image with continuous strokes.

[0084] Image enhancement processing module 20 is used to input the binarized image into a convolutional neural network, extract the skeleton topological structure features of the characters, extract stroke width uniformity parameters and substructure unit spacing consistency parameters from the skeleton topological structure features, compare the skeleton topological structure features with standard character templates, and generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include structural deviation types and the position information of the structural deviation types in the handwritten character image.

[0085] The feature extraction and analysis module 30 is used to extract the pen tip movement trajectory during the writing process using optical flow. Based on the trajectory point sequence marked with continuous timestamps, it analyzes the changes in movement direction and instantaneous speed between adjacent trajectory points. Based on the geometric features of the writing motion, it assigns a higher evaluation weight to high curvature trajectory segments than to straight trajectory segments through a dynamic temporal feature analysis step. It analyzes the writing sequence standardization and pen rhythm consistency through temporal matching to obtain temporal standardization index and temporal deviation analysis results. The temporal deviation analysis results include the temporal deviation type and the position information of the temporal deviation type in the writing sequence.

[0086] The comprehensive scoring module 40 is used to perform weighted fusion of the structural standardization index and the temporal standardization index according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result and the temporal deviation analysis result are input into the scoring model trained by machine learning to calculate the standardization score.

[0087] The error correction report generation module 50 is used to automatically determine the error correction intensity and guidance content based on the level of the standardization score, the structural deviation analysis result and the temporal deviation analysis result, and generate and output a hierarchical error correction report. The hierarchical error correction report includes a targeted copying template and a demonstration of the writing order.

[0088] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0090] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A handwriting standardization monitoring method based on image enhancement, characterized in that, Includes the following steps: The original image of the handwritten character is acquired by an image acquisition device equipped with multi-source adaptive lighting function. Geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation are performed on the original image of the handwritten character in sequence to obtain a binary image with continuous strokes. The binarized image is input into a convolutional neural network to extract the skeleton topological features of the characters. Stroke width uniformity parameters and substructure unit spacing consistency parameters are extracted from the skeleton topological features. The skeleton topological features are compared with standard character templates to generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include the structural deviation type and the position information of the structural deviation type in the handwritten character image. The pen tip movement trajectory during the writing process is extracted using optical flow. Based on a sequence of trajectory points marked with continuous timestamps, the changes in motion direction and instantaneous velocity between adjacent trajectory points are analyzed. Furthermore, based on the geometric characteristics of the writing motion, a dynamic temporal feature analysis step assigns a higher evaluation weight to high-curvature trajectory segments than to straight trajectory segments. Temporal matching analysis is used to determine the consistency of writing sequence regularity and pen stroke rhythm, yielding temporal regularity indices and temporal deviation analysis results, including: Based on the continuously extracted pen tip motion trajectory using the optical flow method, a sequence of trajectory points marked by continuous timestamps is obtained; The ratio of coordinate displacement to time interval between adjacent trajectory points is continuously calculated to obtain the change in motion direction and instantaneous velocity, and the direction change sequence and velocity change sequence are updated in real time. Based on the statistical characteristics of the writing direction change, the high curvature region is detected and marked on the input direction change sequence. The high curvature region is defined as a continuous trajectory segment where the direction change exceeds a preset direction threshold. Based on the state of the detected high curvature region, an evaluation weight coefficient is dynamically assigned to the trajectory segment. The weight coefficient includes a first evaluation weight coefficient assigned to the trajectory segment in the high curvature region and a second evaluation weight coefficient assigned to the trajectory segment in the non-high curvature region. The first evaluation weight coefficient is greater than the second evaluation weight coefficient. The trajectory point sequence is weighted based on the first evaluation weight coefficient and the second evaluation weight coefficient to obtain a weighted trajectory point sequence. Based on the weighted trajectory point sequence, the minimum cumulative path distance between the pen tip movement trajectory and the standard writing order time sequence information in the standard character template is calculated by the dynamic time warping algorithm, and the minimum cumulative path distance is mapped to the trajectory order deviation. By comparing the velocity of the weighted trajectory point sequence with the distribution difference of the standard velocity curve in the standard writing sequence time information, a velocity consistency index is calculated. Based on the trajectory sequence deviation, the speed consistency index, the position information of the abnormal trajectory segment corresponding to the trajectory sequence deviation in the pen tip movement trajectory, and the position information of the abnormal speed segment corresponding to the speed consistency index in the pen tip movement trajectory, a temporal regularity index and a temporal deviation analysis result are generated. The results of the temporal deviation analysis include the type of temporal deviation and the position information of the type of temporal deviation in the written sequence; The structural standardization index and the temporal standardization index are weighted and fused according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result, and the temporal deviation analysis result, are input into a scoring model trained by machine learning to calculate the standardization score. Based on the standardization score, the structural deviation analysis results, and the temporal deviation analysis results, the error correction intensity and guidance content are automatically determined, and a tiered error correction report is generated and output. The tiered error correction report includes targeted copying templates and writing order demonstrations.

2. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, The morphological opening operation uses a 5×5 rectangular structural element.

3. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, The image acquisition device includes: An image sensor with 24-bit analog-to-digital conversion capability is used to improve the color depth and brightness resolution of the original image of handwritten characters; The light source module with adaptive supplemental lighting function is used to automatically adjust the supplemental lighting brightness according to the ambient light intensity in order to suppress paper reflection and shadow occlusion.

4. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, The standard character template includes standard skeleton topology information and standard writing order timing information for each character. The standard writing order timing information includes the writing order, direction, and standard speed curve of each stroke.

5. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, The process involves inputting the binarized image into a convolutional neural network to extract the skeleton topological features of the characters. From these features, stroke width uniformity parameters and sub-structural unit spacing consistency parameters are extracted. The skeleton topological features are then compared with a standard character template to generate structural regularity indicators and structural deviation analysis results, including: By using a convolutional neural network based on the ResNet-50 architecture, the spatial coordinates of key points of character strokes in the binarized image are regressed, and the spatial coordinates of the key points are used to construct the skeleton topological features, wherein the key points include stroke endpoints, intersections, and high curvature points. Based on the spatial coordinates of the key points, adjacent key points are connected by a linear interpolation algorithm to generate continuous stroke center lines. Based on the character structure specification, key points that are spatially adjacent and topologically related are clustered into sub-structural units to obtain sub-structural unit segmentation information; Based on the stroke center line, the variance of the width of the connected pixel regions on both sides of its normal direction is calculated to obtain the stroke width uniformity parameter. Based on the sub-structural unit segmentation information, the Euclidean distance between the center points of the minimum bounding rectangle of the sub-structural unit is calculated and compared with the standard spacing of the corresponding sub-structural unit in the standard character template. The relative deviation value is calculated to obtain the sub-structural unit spacing consistency parameter. Calculate the Hausdorff distance between the skeleton topological features and the corresponding keypoint set in the standard character template to generate a structural deviation matrix; The stroke width uniformity parameter, the substructure unit spacing consistency parameter, and the structural deviation matrix are weighted and summed to generate a structural standardization index; based on the deviation values ​​and position information of each key point in the structural deviation matrix, structural deviation analysis results are generated.

6. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, The scoring model trained based on machine learning calculates a prescriptive score, including: The comprehensive normative value, stroke width uniformity parameter, and substructure unit spacing consistency parameter are normalized to generate a feature vector with uniform dimensions. The positional information in the structural deviation analysis results and the temporal deviation analysis results is encoded into a spatial weight matrix, and the spatial weight matrix corresponds one-to-one with the spatial position of the strokes and sub-structural units in the handwritten character image; The scoring model trained by the machine learning is a prescriptive scoring model based on a style transfer network. The feature vector is forward propagated through the prescriptive scoring model to output a global stroke thickness uniformity score and a global substructure unit spacing coordination score. Based on the spatial weight matrix, the global stroke thickness uniformity score is decomposed and mapped to the corresponding position of each stroke. The local uniformity score of each stroke is obtained by combining the spatial weights corresponding to each position. Based on the spatial weight matrix, the global substructure unit spacing coordination score is decomposed and mapped to the corresponding position of each substructure unit. The local coordination score of each substructure unit is obtained by combining the spatial weights corresponding to each position. The local uniformity scores of all strokes and the local coordination scores of all substructural units are weighted and summed, and the summation results are mapped to a scale from level 1 to level 10 to generate the final normalization score.

7. The handwriting standardization monitoring method based on image enhancement according to claim 1, characterized in that, Based on the normative scoring level, the structural deviation analysis results, and the temporal deviation analysis results, the system automatically determines the error correction intensity and guidance content, generates and outputs a tiered error correction report, including: The normative score, the structural deviation type from the structural deviation analysis results, and the temporal deviation type from the temporal deviation analysis results are all input into a predefined mapping rule. The mapping rule defines the error correction intensity level corresponding to the combination of normative scoring level and deviation type based on the scoring interval division, so as to output the error correction intensity level of high intensity, medium intensity or low intensity. Based on the deviation position coordinates in the structural deviation analysis results, standard images of corresponding strokes or sub-structural units are extracted from the standard character template to generate the first copying template. Based on the abnormal trajectory segment position information in the temporal deviation analysis results, the standard writing order temporal information of the corresponding stroke is called from the standard character template to generate the first writing order demonstration animation; Based on the error correction intensity level, the first copy template and the first writing order demonstration animation are differentiated to obtain an enhanced copy template and writing order demonstration animation; The differentiated processing includes, when the error correction intensity level is high, superimposing visual guidance on the pen stroke direction and force on the first copying template and reducing the playback speed of the first writing order demonstration animation; when the error correction intensity level is medium, superimposing visual guidance on the pen stroke direction on the first copying template and appropriately reducing the playback speed of the first writing order demonstration animation; and when the error correction intensity level is low, directly outputting the first copying template and the first writing order demonstration animation as the basic content. The enhanced copying template and writing order demonstration animation are combined with the standardization score, the structural deviation analysis results, and the temporal deviation analysis results to generate the hierarchical error correction report.

8. A handwriting standardization monitoring system based on image enhancement, characterized in that, The system is used to implement the image-enhanced handwriting standardization monitoring method according to any one of claims 1 to 7, and the system comprises: The image acquisition module is used to acquire the original image of the handwritten character through an image acquisition device equipped with multi-source adaptive lighting function, and to sequentially perform geometric distortion correction, adaptive binarization processing based on local image block brightness statistics, and morphological opening operation on the original image of the handwritten character to obtain a binary image with continuous strokes. An image enhancement processing module is used to input the binarized image into a convolutional neural network, extract the skeleton topological features of the characters, extract stroke width uniformity parameters and substructure unit spacing consistency parameters from the skeleton topological features, compare the skeleton topological features with standard character templates, and generate structural standardization indicators and structural deviation analysis results. The structural deviation analysis results include structural deviation types and the location information of the structural deviation types in the handwritten character image. The feature extraction and analysis module is used to extract the pen tip movement trajectory during the writing process using optical flow. Based on the trajectory point sequence marked with continuous timestamps, it analyzes the changes in movement direction and instantaneous speed between adjacent trajectory points. Based on the geometric features of the writing motion, it assigns a higher evaluation weight to high-curvature trajectory segments than to straight trajectory segments through a dynamic temporal feature analysis step. It also analyzes the writing sequence standardization and pen rhythm consistency through temporal matching to obtain temporal standardization index and temporal deviation analysis results. The temporal deviation analysis results include the type of temporal deviation and the position information of the type of temporal deviation in the writing sequence. The comprehensive scoring module is used to weight and fuse the structural standardization index and the temporal standardization index according to preset weights to obtain a comprehensive standardization value. The comprehensive standardization value, along with the stroke width uniformity parameter, the substructure unit spacing consistency parameter, the structural deviation analysis result, and the temporal deviation analysis result, are input into a scoring model trained based on machine learning to calculate the standardization score. The error correction report generation module is used to automatically determine the error correction intensity and guidance content based on the standardization score level, the structural deviation analysis results and the temporal deviation analysis results, and generate and output a hierarchical error correction report. The hierarchical error correction report includes a targeted copying template and a demonstration of the writing order.

Images