A 3D stereoscopic display method for electronic ink screen

Abstract

The present application relates to a 3D stereoscopic display method for an electronic ink screen, comprising the following steps: analyzing the effects of different grating widths and reflection angles on screen display, dividing the screen into multiple display areas and obtaining an optimized area division scheme according to the effects of different grating widths and reflection angles on screen display, determining the mapping relationship between each display area and 3D image information according to the area division scheme, and further obtaining depth mapping basic data, generating a standard distribution of content types corresponding to each area according to the area division scheme and the depth mapping basic data, using a genetic algorithm, outputting an optimal area combination scheme according to the content type and its priority to the grating width and reflection angle, and using the area division scheme and the depth mapping basic data to adjust the content or modify the grating parameters according to the grating width and reflection angle of each display area. The present application can improve the 3D stereoscopic display effect of the electronic ink screen.

Description

Technical Field

[0001] The present application relates to the technical field of electronic ink screens, and in particular to a 3D stereoscopic display method of electronic ink screens. Background Art

[0002] When realizing high-resolution 3D electronic ink screen display, there is a technical contradiction between grating width and stereoscopic display effect. On the one hand, in order to present more delicate image details, the grating width needs to be as small as possible, which can improve the display resolution. But on the other hand, too small grating width may limit the reflection angle of light, thus affecting the depth and realism of 3D stereoscopic display.

[0003] The electronic ink screen has been divided into multiple areas during the production process, and each area has fixed grating width and reflection angle parameters. In practical applications, it is necessary to select the most appropriate area combination according to the 3D image content to be displayed, which requires in-depth analysis of the display requirements of different types of 3D content, and at the same time, comprehensively consider the differences in grating width and reflection angle of each area. Therefore, how to find an optimal area combination solution based on the existing electronic ink screen partitions to improve the 3D stereoscopic display effect of the electronic ink screen is a technical problem that technicians in this field need to solve urgently. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the purpose of this application is to provide an electronic ink screen 3D stereoscopic display method.

[0005] The electronic ink screen 3D stereoscopic display method described in the present application comprises the following steps:

[0006] Analyze the impact: Analyze the grating width and reflection angle of the electronic ink screen, including measuring and recording the light reflection characteristics under different grating widths, and the impact of the reflection angle on image clarity and contrast, to obtain the impact of different grating widths and reflection angles on screen display;

[0007] Area division: According to the effects of different grating widths and reflection angles on screen display, the screen is divided into multiple display areas, classified according to grating width and reflection angle, and each area is matched with the corresponding grating width and reflection angle values ​​according to the aspect ratio of the screen to obtain an area division plan optimized for the screen;

[0008] Establishing mapping relationship: According to the area division scheme, determine the mapping relationship between each display area and the 3D image information, obtain the display effect of each area at different depth layers according to the mapping relationship, obtain the expected display parameters, determine the stereoscopic display effect of the screen, and optimize the stereoscopic depth of the image to obtain the depth mapping basic data;

[0009] Adjust content distribution: Generate a standard distribution of content types for each region based on the regional division plan and deep mapping basic data. Combine real-time data and user feedback to adjust the content distribution of each region so that the content types displayed in each region achieve the expected effect.

[0010] Output combination scheme: Genetic algorithm is used to output the optimal regional combination scheme according to the content type and its priority for grating width and reflection angle;

[0011] Adjust display content or parameters: Utilize the area division scheme and depth mapping basic data, combined with user feedback and real-time monitoring data, adopt the random forest model and fuzzy logic control system, establish the correlation model between content type and electronic ink screen display effect, monitor the actual display results on the electronic ink screen in real time, and compare and analyze with the expected results, adjust the content or modify the grating parameters according to the grating width and reflection angle of each display area.

[0012] Preferably, the step of analyzing the impact effect is specifically as follows:

[0013] According to different grating widths and materials, the reflection angle, absorptivity, transmittance and refractive index of light on the grating surface are measured and recorded to obtain the light reflection characteristic data under different grating widths and material combinations;

[0014] Analyze the reflection angle distribution under different grating parameters, use light under different reflection angles to illuminate the electronic ink screen, collect the screen display image through the image sensor, and quantitatively evaluate the image clarity and contrast to obtain the effect data of different reflection angles on the image quality of the screen display;

[0015] A mathematical model is established based on the data between grating parameters, reflection angle and image quality. The grating width, material and reflection angle are used as input features and the image quality evaluation index is used as the output target. A regression model is trained to analyze the effects of different grating widths and reflection angles on screen display.

[0016] Preferably, the area division step is specifically as follows:

[0017] According to the effects of different grating widths and reflection angles on the display effect of the electronic ink screen, experimental data are obtained through optical testing, including the reflected light intensity, reflection angle distribution, contrast, and color saturation under different grating parameter combinations, and a data set between grating parameters and display effects is constructed;

[0018] Data preprocessing technology is used to clean, normalize and select features of the acquired optical characteristic data, remove noise data and redundant features, evaluate the importance of features through filtering or wrapping methods, select key features that best reflect the relationship between grating parameters and display effects, and construct an optimized data set;

[0019] The optimized data set is input into the K-means clustering algorithm. By calculating the clustering performance index under different K values, the K value with the best performance and the appropriate number of clusters is selected in combination with the elbow rule or the Silhouette rule. Cluster analysis is performed to divide the display effect data points under different grating parameter combinations into K cluster centers. Each cluster represents a typical display effect mode, and the display area division scheme of the electronic ink screen is obtained.

[0020] For each display area, the optimal grating width and reflection angle parameter range corresponding to the area is determined according to the cluster center to which it belongs, and a nonlinear mapping relationship between the grating parameters and the display effect in the area is established to obtain the grating parameter optimization strategy for each display area;

[0021] Taking into account the aspect ratio and resolution of the electronic ink screen, the screen is divided into several grid units, each of which corresponds to a display area.

[0022] According to the position and size of each grid unit, the corresponding grating width and reflection angle parameters are calculated to generate a regional parameter mapping table;

[0023] According to the mapping table, the optimal grating parameter combination is assigned to each grid unit, and additional parameter adjustment and optimization are performed on the grid units in the edge area of ​​the screen to generate a regional layout map that is compatible with the screen specifications;

[0024] Furthermore, according to the effects of different grating widths and reflection angles on screen display, each area is matched with an image with predetermined contrast content or predetermined detail richness. The image with predetermined contrast content or predetermined detail richness will be allocated to a predetermined display area, and other content will be allocated to other areas.

[0025] Preferably, according to the effects of different grating widths and reflection angles on screen display, each area is matched with an image with a predetermined contrast content or a predetermined detail richness, and an image with a predetermined contrast content or a predetermined detail richness will be allocated to a predetermined display area, and other content will be allocated to other areas. Specifically:

[0026] According to the display area division scheme of the screen, the grating width and reflection angle of each area are analyzed, and the mapping relationship between regional grating parameters and display effects is established. Through optical simulation and measured data, the display effect evaluation index of different areas under different grating parameters is obtained;

[0027] According to the usage scenario and user needs, key display areas are pre-set as the scheduled display areas;

[0028] An image processing algorithm is used to analyze the image content to be displayed, extract the image contrast information and detail richness information, and obtain the display effect evaluation index of the image content;

[0029] For contrast, the global contrast and local contrast of the image are calculated, measured by the grayscale standard deviation and gradient amplitude mean of the image, respectively. According to the preset global and local contrast thresholds, the image is divided into high contrast and low contrast categories;

[0030] For detail richness, the edge density and texture complexity of the image are calculated, and the edge pixel ratio obtained by the Canny edge detection algorithm and the entropy of the texture histogram obtained by the local binary pattern operator are used to measure them respectively. According to the preset edge density and texture complexity thresholds, the image is divided into two categories: high detail and low detail, and the classification labels of the image content are obtained. The number and proportion of each category of image content are counted;

[0031] Matching the classification label of the image content with the display effect evaluation index of the display area, and finding the optimal image content, including a matching scheme of a predetermined contrast content or a predetermined detail richness image and the display area through the Hungarian algorithm or the KM algorithm;

[0032] During the matching process, a higher priority is assigned to the predetermined display area, and image contents with high contrast and high details are preferentially assigned to the predetermined display area;

[0033] For non-predetermined display areas, corresponding image content is allocated according to the quality of their display effects;

[0034] According to the matching results between the image content and the display area, a multivariate nonlinear regression model is constructed, with the grating width and the reflection angle as independent variables and the display effect evaluation index as the dependent variable, and the least square method or gradient descent method is used to fit the model parameters;

[0035] An optimization objective function is constructed for each region. The objective function is composed of the difference between the display effect evaluation index predicted by the regression model and the matched image content characteristics and the regularization term of the grating parameters. The optimal grating parameter combination of each region is solved by the gradient descent method or the quasi-Newton method so that the display effect of each region is adapted to the characteristics of the matched image content.

[0036] According to the extracted image content features, the classification labels of the image content are updated in real time, and the region matching algorithm and grating parameter optimization algorithm are executed. An incremental update strategy is adopted to reclassify and match only the image content that has changed significantly.

[0037] Preferably, the step of establishing a mapping relationship is specifically as follows:

[0038] According to the physical characteristics and display requirements of the electronic ink screen, the screen is divided into multiple display areas using a layered division method, each area has a different grating width and reflection angle;

[0039] For the input 3D image, a stereo matching algorithm is used to match the corresponding points of the left and right views, calculate the disparity value of the pixel points, and then triangulate the depth value based on the disparity and camera parameters to obtain the depth map;

[0040] According to the display area of ​​the screen, the depth map is divided into blocks and quantized to obtain the pixel distribution of each area at different depth layers;

[0041] Based on the display area division and depth layer division, the optical characteristics of each area at different depth layers are modeled to establish a three-dimensional data table of area-depth-optical characteristics;

[0042] According to the expected effects of 3D stereoscopic display, including depth of field range, parallax size and interlacing effect, the optical characteristics of each area at different depth layers are quantitatively analyzed and evaluated to obtain a set of objective indicators for measuring the display effect;

[0043] The area-depth-optical characteristic data table is used as a training sample, and the display effect evaluation index is used as a training target. A three-dimensional mapping model is obtained through training to predict the display effect of each area at different depth layers.

[0044] According to the importance and weight of the display effect index, the three-dimensional mapping model is optimized and simplified to obtain a simplified mapping relationship expression or mapping function for real-time calculation of the expected display parameters of each area at different depth layers, wherein the expected display parameters include grating width and reflection angle;

[0045] The grating parameter optimization problem is modeled as a multi-objective constrained optimization problem, with the expected display parameters as the optimization target and the mapping relationship between the display area and the depth layer as the constraint condition, and the objective function and constraint equation of the optimization model are established;

[0046] According to the characteristics of the objective function and constraint equations, a suitable optimization algorithm is selected, and according to the characteristics and parameter settings of the selected optimization algorithm, the optimization model is solved to obtain the global optimal or local optimal grating parameter combination;

[0047] The optimized grating parameter combination is simulated and tested in real objects to evaluate its actual display effect at different depth layers until the actual display effect meets the display effect requirements;

[0048] Furthermore, the display effect of the predetermined layer information of each area is analyzed to obtain the display performance of each area at different depth layers, and the area suitable for displaying the first predetermined depth layer and the area suitable for displaying the second predetermined depth layer are analyzed and determined.

[0049] Preferably, the display effect of the predetermined layer information of each area is analyzed to obtain the display performance of each area at different depth layers, and the area suitable for displaying the first predetermined depth layer and the area suitable for displaying the second predetermined depth layer are analyzed and determined specifically as follows:

[0050] According to the resolution and size of the electronic ink screen, the screen is divided into several display areas by using an equidistant grid division method, and the size and number of pixels of each display area are the same;

[0051] For each display area, determine the first predetermined depth layer information and the second predetermined depth layer information corresponding to the area according to its location and size, obtain the optical characteristic parameters of the area at different depth layers through optical simulation and experimental measurement, including brightness, contrast and color saturation, and establish a three-dimensional mapping relationship between area-depth layer-optical characteristic;

[0052] Continuously acquire images of the display performance process of each display area at different depth layers to obtain a dynamic image sequence, and then use an image processing algorithm to calculate the dynamic characteristic parameters of each display area, wherein the dynamic characteristic parameters include the response time of the display state and the grayscale change curve;

[0053] At the same time, an optical sensor is used to measure and obtain the optical characteristic parameters of the display performance picture of each display area at different depth layers, which are recorded as static characteristic parameters;

[0054] The dynamic characteristic parameters and the static characteristic parameters are associated with the depth layer to construct a three-dimensional data matrix of region-depth-display characteristics. Based on the obtained three-dimensional data matrix, the distribution law and influencing factors of the display characteristics at different depth layers are obtained;

[0055] The display characteristic parameters of each region at different depth layers are taken as the depth attribute vector of the region, and then a clustering algorithm is used to cluster all regions into several region groups with similar depth attributes. Each region group corresponds to an optimal depth layer, that is, the display characteristics of all regions in the region group at this depth layer are optimal.

[0056] Preferably, the step of adjusting content distribution includes:

[0057] According to the area division scheme and the depth mapping basic data, the depth features of each display area are analyzed, and the depth distribution histogram and depth gradient histogram of the area are extracted as feature vectors. In combination with the predefined content type label system, the regional display data, including the depth features and content type labels of the area, are collected to construct a training data set;

[0058] Preprocess and feature engineer deep features to extract the most discriminative feature subset;

[0059] Select a classification algorithm based on data size, feature type, and computing resources;

[0060] Cross-validation and grid search methods are used to optimize the algorithm's hyperparameters to improve classification performance;

[0061] Evaluate the accuracy, precision, and recall of the classification model on the test set, compare the performance of different algorithms, select the optimal model, train the region-depth-content type classification model, and obtain the standard distribution of content types suitable for display in each region;

[0062] Using data mining algorithms, we can discover the association rules and co-occurrence patterns between different regions, different depth layers, and different content types from historical data, and generate association rule libraries and frequent pattern libraries for regional content distribution as the basis for content distribution and recommendation.

[0063] Collect multi-dimensional attribute data of users and construct user feature vectors;

[0064] Non-dimensionalize user features and select a user similarity measurement method;

[0065] Use clustering algorithms to group users, and determine the optimal number of clusters and clustering results based on clustering effects and business interpretability;

[0066] Conduct statistical analysis on the content preferences of each user group and construct a group-content type preference matrix;

[0067] Build a user portrait model, divide users into multiple user groups, and model the content preferences of each user group;

[0068] Collect user behavior data in real time and recommend content based on the behavior data;

[0069] Conduct comparative tests on different regional content distribution plans, evaluate the display effects of different plans, and select the optimal content distribution plan.

[0070] Preferably, the step of outputting the combination scheme is specifically as follows:

[0071] According to the display area division of the electronic ink screen and the content display requirements, different types of content are classified and labeled, a correspondence matrix between content type and display area is established, and the grating width and reflection angle parameters of each content type are initialized as optimization variables of the genetic algorithm;

[0072] For each display area, according to the type of content to be displayed, the value range and constraints of the grating width and reflection angle parameters are defined, and the objective function of parameter optimization is established. The indicators of display quality and energy consumption are comprehensively considered and quantified by weighted summation. The weight coefficient is adjusted according to actual needs and preferences;

[0073] Using real number coding, the grating width and reflection angle parameters of each display area are mapped to the individual genes of the genetic algorithm. The initial population is generated by random sampling, and the fitness function value of each individual is calculated according to the parameter value of the individual gene, and the population is sorted.

[0074] Perform genetic manipulation on individuals in the population, giving priority to individuals with high fitness for reproduction;

[0075] According to the priority of the content type, the individual genes are modified and adjusted, and a priority weight matrix is ​​set to assign different weight coefficients to different content types. When evaluating fitness, the weight coefficients are multiplied with the display quality and energy consumption indicators to obtain the weighted fitness value, thereby realizing the preference for high-priority content types.

[0076] For gene positions corresponding to high-priority content types, reduce their crossover and mutation probabilities to maintain excellent characteristics;

[0077] For gene positions corresponding to low-priority content types, their crossover and mutation probabilities are increased to speed up elimination and updating;

[0078] For individuals that violate the priority constraints, a penalty term is imposed on the fitness value to reduce their probability of being selected;

[0079] Recalculate the fitness function value of the new generation population after genetic operation, and retain the elite and merge the population with the previous generation population to control the population size to be stable within a certain range;

[0080] In addition, according to the requirements of smoothness and clarity of the displayed content, the corresponding grating width and reflection angle are matched. Through a genetic algorithm, dynamic video content is prioritized in areas where the grating width and reflection angle are balanced, and static graphic content is placed in areas where the grating is narrow and the reflection angle is predetermined.

[0081] Preferably, according to the requirements of the smoothness and clarity of the displayed content, the corresponding grating width and reflection angle are matched, and the dynamic video content is preferentially arranged in the area where the grating width and reflection angle are balanced through the genetic algorithm, and the static chart content is placed in the area where the grating is narrow and the reflection angle is predetermined. Specifically:

[0082] According to the display area division of the electronic ink screen, different types of display content are identified and classified. The image classification algorithm based on deep learning is used to extract features and judge the type of input display content, and obtain different types of content labels such as dynamic videos, static charts, and texts for subsequent area matching and optimization.

[0083] For each type of display content, a mapping relationship model between the content type and the grating width and reflection angle is established according to the requirements of its fluency and clarity;

[0084] Construct an electronic ink screen display content dataset, including display samples of different types and parameter configurations, as well as corresponding evaluation indicators of fluency and clarity, as a training dataset;

[0085] For the fluency index, frame rate, frame delay, and frame loss are used as objective indicators, and user subjective ratings are used as subjective indicators;

[0086] For clarity index, contrast, sharpness, and color saturation are used as objective indicators, and user subjective rating is used as subjective indicator;

[0087] A deep learning algorithm is used to train and obtain a mapping model by taking the image features and grating parameter configuration of the displayed content as input and the fluency and clarity indicators as output;

[0088] For dynamic video content, by analyzing its frame rate, motion speed, scene complexity, and matching the appropriate grating width and reflection angle parameter range, the smoothness and continuity of video playback are ensured;

[0089] Genetic algorithm is used to optimize the parameter combination, and the average frame rate is taken as the primary optimization target of the genetic algorithm;

[0090] For static chart content, by analyzing its edges, textures and colors, the appropriate grating width and reflection angle parameter range are matched to ensure the clarity and readability of the chart display;

[0091] Genetic algorithm is used to optimize the parameter combination, and contrast is taken as the primary optimization target;

[0092] For the text content, heuristic rules or search algorithms based on constraint satisfaction are used to allocate it to the remaining display area, and the grating width and reflection angle parameters are fine-tuned according to its fluency and clarity requirements.

[0093] Preferably, the step of adjusting the display content or parameters is specifically:

[0094] According to the physical characteristics and display requirements of the electronic ink screen, an adaptive partitioning algorithm is used to divide the screen into multiple rectangular display areas;

[0095] Based on the semantic segmentation information of the displayed content, a deep learning model is used to perform pixel-level semantic annotation on the displayed image, identify different content types and display objects, and generate a semantic segmentation map, wherein the content types include text, images, and videos, and the display objects include titles, text, and icons;

[0096] Combining the semantic segmentation map and pixel density information, the regions with similar semantics and similar pixel density are merged into one display area;

[0097] The divided display areas are encoded and annotated to generate a region division map, and the position, size, pixel density and semantic category of each display area are extracted. Based on the extracted information, the parameters of each display area are optimized and dynamically adjusted.

[0098] The advantages of the 3D display method of an electronic ink screen described in this application are:

[0099] This application solves the problem of poor display effect of electronic ink screen in the prior art by analyzing the grating width and reflection angle of the electronic ink screen. First, the influence of different grating widths and reflection angles on screen display is measured and recorded to obtain the influence of grating parameters on image clarity and contrast. Then, the K-means clustering algorithm is used to divide the screen into multiple display areas, and the corresponding grating width and reflection angle are matched to each area to form an optimized area division scheme.

[0100] Furthermore, the present application matches the display content of each area according to the image with a predetermined contrast or detail richness, so that the image is displayed in its respective optimal area, optimizes the stereoscopic depth of the image by determining the mapping relationship between the display area and the 3D image information, obtains the basic data of the depth mapping, and optimizes the display effect according to different depth layers. In addition, based on the depth mapping data and the area division scheme, a standard distribution of content types is generated, and combined with real-time data and user feedback, the display content is dynamically adjusted to ensure that the display effect of each area meets the expectations.

[0101] This application also uses a genetic algorithm to output the optimal regional combination plan based on the content type and its priority for grating parameters, so that dynamic video content and static graphic content can be optimally displayed respectively. Through the random forest model and fuzzy logic control system, a correlation model between the content type and the electronic ink screen display effect is established, and the display results are monitored and adjusted in real time to ensure that the actual display effect is consistent with expectations. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 is a flowchart of steps for optimizing the display effect in this embodiment;

[0103] Figure 2 is a flowchart of the steps for optimizing display parameters in this embodiment;

[0104] Figure 3 is a flow chart of the steps of quantizing pixel distribution of the depth map in this embodiment. DETAILED DESCRIPTION

[0105] like Figures 1 to 3 As shown, the electronic ink screen 3D stereoscopic display method described in the present application includes the following steps:

[0106] Analyze the grating width and reflection angle of the electronic ink screen, including measuring and recording the light reflection characteristics under different grating widths, and the influence of the reflection angle on image clarity and contrast, and obtain the influence of different grating widths and reflection angles on screen display.

[0107] According to different grating widths and materials, the optical parameters such as the reflection angle, absorptivity, transmittance and refractive index of the light on the grating surface are measured and recorded to obtain the light reflection characteristic data under different grating widths and material combinations. Theoretical analysis or numerical simulation is performed through ray tracing and other methods to analyze the distribution of reflection angles under different grating parameters, providing a basis for the selection of reflection angles. Light at different reflection angles is used to illuminate the electronic ink screen, and the screen display image is collected through the image sensor. The quality evaluation indicators such as image clarity and contrast are quantitatively evaluated to obtain the effect data of different reflection angles on image quality. According to the data between grating parameters, reflection angle and image quality, a mathematical model is established. A supervised learning algorithm such as support vector machine (SVM) or neural network is used to train the regression model with grating width, material and reflection angle as input features and image quality evaluation indicators as output targets. The model hyperparameters are optimized through grid search or random search. Taking the best image quality as the optimization goal, grating width, material and light reflection angle as optimization variables, heuristic optimization algorithms such as genetic algorithm and particle swarm optimization algorithm are used to search for the optimal grating design parameter combination of electronic ink screen through model prediction and simulation.

[0108] Exemplarily, when measuring the optical parameters of the grating surface, a spectrophotometer can be used to measure the reflectivity and transmittance at different incident angles, and an ellipsometer can be used to measure the light reflection characteristic data of the grating surface. For example, for a grating with a width of 200nm and a material of ITO, the reflectivity is measured to be 20%, the transmittance is 75%, and the refractive index is 2.1 at an incident angle of 30°. Using the measured optical parameters, ray tracing analysis is performed through optical simulation software such as TracePro to simulate the reflection angle distribution under different grating parameters. The simulation results show that when the grating width is 200nm and the material is ITO, the reflection angle is mainly concentrated between 20°-40°. When evaluating the effect of reflection angle on image clarity, a CMOS image sensor with a resolution of 1080P is used to capture screen display images at different reflection angles, and the image clarity (such as MTF value) and contrast (such as RMS contrast) are calculated through image processing libraries such as OpenCV. Experimental data show that when the reflection angle is between 20° and 40°, the MTF value and RMS contrast are the highest, reaching 0.85 and 0.6 respectively. Based on the above data, the Python scikit-learn library is used to build an SVM regression model, with grating width, ITO refractive index and reflection angle as input features, MTF value and RMS contrast as output targets, and training and testing using a 5-fold cross validation method. The kernel function and regularization parameters of the SVM are optimized by grid search, and the average absolute error of the optimal model on the test set is 0.02. The trained SVM model is combined with the genetic algorithm for multi-objective optimization, and the weighted sum of the MTF value and RMS contrast is used as the fitness function to search for the optimal combination of grating width, ITO refractive index and reflection angle. The population size of the genetic algorithm is set to 50, the evolutionary generation is 100, the crossover probability is 0.8, and the mutation probability is 0.1. The optimization results show that when the grating width is 180nm, the ITO refractive index is 2.0, and the reflection angle is 35°, the MTF value and RMS contrast reach 0.92 and 0.65 respectively, which is the optimal design parameter combination. According to the optimal parameter combination, a grating structure with a period of 180nm and an aspect ratio of 1:1 is prepared on the surface of the electronic ink screen by electron beam lithography and reactive ion etching, and a 100nm thick ITO film is deposited on the grating surface by magnetron sputtering to obtain the optimized electronic ink screen sample. The optical performance of the sample is tested. At an incident angle of 35°, the MTF value and RMS contrast are 0.90 and 0.63 respectively, which are basically consistent with the optimization target, proving the effectiveness of the optimized design. At the same time, compared with the unoptimized electronic ink screen, the image quality is improved by more than 20%, which fully demonstrates the significant effect of grating optimization.

[0109] The K-means clustering algorithm is used to divide the screen into multiple display areas according to the effects of different grating widths and reflection angles on screen display. The areas are classified according to the grating widths and reflection angles. According to the aspect ratio of the screen, each area is matched with the corresponding grating width and reflection angle values ​​to obtain an area division scheme optimized for the screen.

[0110] According to the effects of different grating widths and reflection angles on the display effect of the electronic ink screen, a large amount of experimental data is obtained through optical testing, including indicators such as reflected light intensity, reflection angle distribution, contrast, and color saturation under different grating parameter combinations, and a data set between grating parameters and display effects is constructed. Data preprocessing technology is used to clean, normalize, and select features for the acquired optical characteristic data, remove noise data and redundant features, evaluate the importance of features through filtering or wrapping methods, select key features that best reflect the relationship between grating parameters and display effects, and construct an optimized data set to prepare for subsequent clustering analysis. The optimized data set is input into the K-means clustering algorithm, and by calculating clustering performance indicators under different K values, such as silhouette coefficient, Davies-Bouldin index, etc., combined with the elbow rule or Silhouette rule, the K value with the best performance and the appropriate number of clusters is selected for clustering analysis, and the display effect data points under different grating parameter combinations are divided into K cluster centers. Each cluster represents a typical display effect mode, and the display area division scheme of the electronic ink screen is obtained. For each display area, the optimal grating width and reflection angle parameter range corresponding to the area is determined according to the cluster center to which it belongs. Models such as polynomial regression, Gaussian process regression, multivariate linear regression, and partial least squares regression are used. By introducing high-order terms, kernel functions, or feature conversion methods, a nonlinear mapping relationship between grating parameters and display effects in the area is established to obtain the grating parameter optimization strategy for each display area. Considering the aspect ratio and resolution of the electronic ink screen, the screen is divided into several grid units, each of which corresponds to a display area. According to the position and size of each grid unit, the corresponding grating width and reflection angle parameters are calculated to generate a regional parameter mapping table. According to the mapping table, the optimal grating parameter combination is assigned to each grid unit, and additional parameter adjustment and optimization are performed on the grid units in the edge area of ​​the screen to generate a regional layout diagram that is compatible with the screen specifications, guiding the subsequent grating array design and manufacturing. Professional optical design software such as ZemaxOpticStudio and VirtualLab are used to build an electronic ink screen optical model including grating structure, electrode layer, liquid crystal layer, etc. Ray tracing, wavefront analysis and other methods are used to simulate the propagation and diffraction behavior of light at different viewing angles, and the intensity and angular distribution of the emitted light are calculated. Based on the simulation results, the display uniformity, viewing angle dependence and other performance indicators of different regional grating parameter combinations are evaluated, and the regional division scheme and parameter mapping strategy are further optimized to balance display performance and manufacturing cost, and obtain the regionalized grating optimization design scheme for electronic ink screens that can be put into mass production.

[0111] For example, in the optical test, a high-speed CMOS camera and a spectrum analyzer with a resolution of 1920×1080 were used to collect the optical characteristics of the electronic ink screen under 200 different grating widths (50nm-500nm) and reflection angles (0°-60°), and an original data set containing 10,000 data points was obtained. The data was cleaned by Python's Pandas library, and the data with more than 20% missing values ​​were removed, and the outliers were corrected. The light intensity, contrast and other features were normalized by combining the maximum and minimum normalization and Z-score normalization. The Pearson correlation coefficient and mutual information index were used to evaluate the correlation between each feature and the display effect, and the top 10 features with the highest correlation were selected as the input of the optimized data set. The silhouette coefficient and Davies-Bouldin index were used to evaluate the clustering performance under different K values ​​(3-20), and the optimal number of clusters was determined to be 8 in combination with the elbow rule. For the display area represented by each cluster center, a polynomial regression is used to fit the mapping relationship between grating width, reflection angle and display effect. The polynomial order is optimized through 5-fold cross-validation and grid search, and a fitting model with an average R² value of 0.95 is obtained. The electronic ink screen is divided into 120×68 grid units, and each unit is determined to belong to a cluster according to its geometric center position, and the optimal grating parameters are calculated according to the cluster mapping model. The optical model of the electronic ink screen is built in Zemax Optic Studio. Through ray tracing and diffraction analysis, the display effect under different viewing angles (0°-80°) is simulated, and the standard deviation of the display uniformity is calculated to be less than 0.05.

[0112] According to the effects of different grating widths and reflection angles on screen display, each area is matched with an image with predetermined contrast content or predetermined detail richness. The image with predetermined contrast content or predetermined detail richness will be allocated to the predetermined display area, and other content will be allocated to other areas.

[0113] According to the display area division scheme of the screen, the grating width and reflection angle of each area are analyzed, and the mapping relationship between the regional grating parameters and the display effect is established. Through optical simulation and measured data, the display effect evaluation indicators of different areas under different grating parameters, such as contrast, detail richness, etc., are obtained. According to the use scenarios and user needs of the display, some key display areas, such as the center of the screen and the edge of the screen, are pre-set as the predetermined display areas. These areas usually need to display the most important and critical image content, such as text, icons, video bodies, etc., and have the highest requirements for display effects. The image processing algorithm is used to analyze the image content to be displayed, extract the image contrast information and detail richness information, and obtain the quantitative description index of the image content. For contrast, the global contrast and local contrast of the image are calculated, and the grayscale standard deviation and gradient amplitude mean of the image are used to measure respectively. According to the preset global and local contrast thresholds, the image is divided into high contrast and low contrast. For detail richness, the edge density and texture complexity of the image are calculated, and the edge pixel ratio obtained by the Canny edge detection algorithm and the entropy of the texture histogram obtained by the local binary pattern (LBP) operator are used to measure them respectively. According to the preset edge density and texture complexity thresholds, the image is divided into two categories: high detail and low detail, and the classification labels of the image content are obtained. The number and proportion of each category of image content are counted. The classification labels of the image content are matched with the display effect evaluation index of the display area, and the optimal image content, including the matching scheme of the predetermined contrast content or the predetermined detail richness image and the display area, is found through optimization algorithms such as the Hungarian algorithm or the KM algorithm. The Hungarian algorithm is suitable for the situation where the number of image content categories and the number of display areas are small, while the KM algorithm is more general and robust, and the appropriate algorithm needs to be selected according to the specific situation. In the matching process, a higher priority is assigned to the predetermined display area, and high-contrast and high-detail image content is preferentially assigned to these areas, and a greater weight is given to the matching edges of the predetermined display area. For the non-predetermined display area, the corresponding image content is assigned according to the quality of its display effect to achieve the balance and optimization of the overall display effect. According to the matching results between the image content and the display area, a multivariate nonlinear regression model is constructed, with the grating width and reflection angle as independent variables and the display effect evaluation index as the dependent variable. The model parameters are fitted using algorithms such as the least squares method or the gradient descent method. An optimization objective function is constructed for each area. The objective function consists of the difference between the display effect index predicted by the regression model and the matched image content characteristics and the regularization term of the grating parameters. The optimal grating parameter combination for each area is solved using optimization algorithms such as the gradient descent method or the quasi-Newton method, so that the display effect of each area is adapted to the characteristics of the matched image content, thereby improving the overall display quality. The image content analysis module is integrated in the display system, and a parallel computing architecture based on FPGA or GPU is used to achieve high-speed, low-latency image content feature extraction.According to the extracted image content features, the classification labels of the image content are updated in real time, and the execution of the region matching algorithm and the grating parameter optimization algorithm is triggered. The incremental update strategy is adopted to reclassify and match only the image content that has changed significantly. When optimizing the grating parameters, the time regularization term is introduced to set the upper and lower limits of the grating parameter adjustment to ensure the smooth transition and stability of the display effect. Through hardware acceleration and algorithm optimization, real-time processing of image content analysis, region matching and grating parameter adjustment is achieved to meet the needs of dynamic adaptive display.

[0114] For example, on an electronic ink screen with a resolution of 2560×1600, the screen is divided into three types of predetermined display areas: the central area, the edge area, and the corner area. The area of ​​the central area accounts for 60%, the edge area accounts for 30%, and the corner area accounts for 10%. Through optical simulation and actual measurement, the contrast and detail richness data under different grating widths (100nm-500nm) and reflection angles (0°-60°) are obtained, and a ternary quadratic polynomial regression model is constructed to fit the relationship between grating parameters and display effects. The determination coefficient R² of the model reaches above 0.95. The image to be displayed is analyzed, and the Canny algorithm is used to extract edges. Images with edge pixels exceeding 10% are classified as high-detail images; the LBP operator is used to extract texture features, and the entropy of the texture histogram is calculated. Images with entropy values ​​exceeding 3.5 are classified as high-detail images; the grayscale standard deviation and gradient amplitude mean are used to calculate global and local contrast, and images with grayscale standard deviation and gradient amplitude mean values ​​exceeding 50 and 20, respectively, are classified as high-contrast images. The KM algorithm is used for regional matching. Images with high details and high contrast are matched to the central area first, and the matching weight is set to 1.5. Images with low details and high contrast are matched to the edge area, and the matching weight is set to 1.2. Images with low details and low contrast are matched to the corner area, and the matching weight is set to 1.0. An objective function is constructed for each area. The objective function includes the mean square error between the display effect predicted by the regression model and the matching image features, as well as the L2 regularization term of the grating width and reflection angle, and the regularization coefficient is set to 0.01. The Adam optimization algorithm is used for solving, with the learning rate set to 0.001, the number of iterations to 100, and the batch size to 32. The image analysis module is implemented on the FPGA, and the Canny algorithm, LBP operator and other algorithms are mapped to the FPGA logic resources using advanced synthesis tools to achieve pipeline parallel computing. The processing delay of each frame of the image is less than 10ms. In the display control module, the grating width and reflection angle of each area are adjusted in real time according to the image analysis results and area matching results. The adjustment range is limited to ±10%, and the parameter change rate between adjacent frames is limited to ±5% to ensure the smoothness and stability of the display effect.

[0115] According to the area division plan, the mapping relationship between each display area and the 3D image information is determined, and the display effect of each area at different depth layers is obtained according to the mapping relationship. The expected display parameters are obtained, the stereoscopic display effect of the screen is determined, and the stereoscopic depth of the image is optimized to obtain the basic data of the depth mapping.

[0116] According to the physical characteristics and display requirements of the electronic ink screen, the screen is divided into multiple display areas by using a hierarchical division method. Each area has different grating width and reflection angle parameters. By adjusting these parameters, the display effects of different areas can be differentiated, providing a basis for 3D stereo display. For the input 3D image, stereo matching algorithms such as Semi-Global Matching (SGM) and GraphCut are used to match the corresponding points of the left and right views, calculate the disparity value of the pixel points, and then triangulate the depth value based on the disparity and camera parameters; for multi-view videos, depth sensors such as structured light or ToF (Time-of-Flight) cameras are used to directly obtain the depth map of the scene. Through methods such as image registration and interpolation, the depth map is aligned and fused with the multi-view video to obtain the depth value of the pixel points at each view; for 3D models, the position and parameters of the virtual camera are set through the rendering engine, and the 3D model is rendered in depth to directly generate a depth map. According to the display area division of the screen, the depth map is divided into blocks and quantized to obtain the pixel distribution of each area at different depth layers. Based on the division of display areas and depth layers, the optical characteristics of each area at different depth layers are modeled, such as brightness, contrast, color, etc., and a three-dimensional data table of area-depth-optical characteristics is established. According to the expected effects of 3D stereoscopic display, such as depth of field range, parallax size, interlacing effect, etc., the optical characteristics of each area at different depth layers are quantitatively analyzed and evaluated to obtain a set of objective indicators for measuring the display effect. Using machine learning algorithms such as support vector regression (SVR) and Gaussian process regression (GPR), with the area-depth-optical characteristics data table as training samples and the display effect index as the training target, a three-dimensional mapping model is trained to predict the display effect of any area at different depth layers. According to the importance and weight of the display effect index, the three-dimensional mapping model is optimized and simplified, such as feature selection, dimension compression, etc., to obtain a simplified mapping relationship expression or mapping function, which is used to calculate the expected display parameters of each area at different depth layers in real time, such as grating width, reflection angle, etc. The grating parameter optimization problem is modeled as a multi-objective constrained optimization problem, with the desired display parameters as the optimization target and the mapping relationship between the display area and the depth layer as the constraint condition, and the objective function and constraint equation of the optimization model are established. According to the characteristics of the objective function and constraint equation, such as continuity, differentiability, multi-peaks, etc., a suitable optimization algorithm is selected, such as genetic algorithm (GA), particle swarm optimization (PSO), simulated annealing algorithm (SA), etc. According to the characteristics of the algorithm and parameter settings, the optimization model is solved to obtain the global optimal or local optimal grating parameter combination.The optimized grating parameter combination is simulated and tested in real objects to evaluate its actual display effect at different depth layers. If there is a large deviation from the expected effect, it is necessary to adjust the objective function weight or constraint of the optimization model and re-optimize the solution until the display effect requirements are met. In the process of grating parameter optimization, the value range and discretization step of the grating width and reflection angle, as well as their sensitivity and monotonicity to the display effect are considered; the continuity and smoothness of grating parameters between different regions and depth layers, and the transition effect between adjacent regions and depth layers; the physical limitations and cost constraints of grating materials and manufacturing processes, and their impact on the grating parameter optimization space. In the process of 3D stereoscopic display, the disparity distribution of the left and right eye views is obtained in real time, and the disparity value of each pixel is calculated through algorithms such as disparity estimation and disparity matching to form a disparity map. According to the physiological characteristics of the human eye and the visual cognitive mechanism, mathematical models of disparity compensation are established, such as the disparity-depth model and the disparity-fusion time model, which are used to describe the quantitative relationship between disparity and stereoscopic perception effect. According to the disparity compensation model and disparity map, the disparity compensation amount of each pixel is calculated, that is, the disparity value that needs to be increased or decreased to achieve the best stereoscopic perception effect. The disparity compensation amount is fused with the depth value of the pixel, the raster parameters of the display area and other information to obtain the position offset of each pixel in the left and right eye views, and then the left and right eye views are displaced and resampled according to the offset to obtain a stereoscopic image after disparity compensation. In the process of disparity compensation, the spatial resolution and temporal resolution of disparity compensation are focused on, as well as their matching with display refresh rate, response time and other parameters; the amplitude and smoothness of disparity compensation, as well as their relationship with factors such as depth of field range and disparity comfort; the computational complexity and real-time performance of the disparity compensation algorithm, as well as their combination with hardware acceleration, parallel optimization and other means. Finally, the 3D stereoscopic display effect is evaluated by collecting users' visual experience feedback through a combination of subjective and objective methods, and using optical measurement equipment to quantitatively analyze indicators such as the stereoscopic depth, parallax range, and brightness uniformity of the display screen. The display effects before and after optimization are compared, and the depth mapping model and grating optimization algorithm are continuously improved. While ensuring the display quality, the computational efficiency and stability of the 3D stereoscopic display are improved, and high-performance, high-realism 3D stereoscopic electronic ink screen display is achieved.

[0117] For example, an electronic ink screen with a size of 800×600 pixels is divided into 3×3 grids to obtain 9 display areas, each with a size of 266×200 pixels. The grating structure of each area is modeled and optimized by the optical simulation software Zemax OpticStudio. Within the range of grating width of 200-400nm and reflection angle of 10°-60°, the parameters are scanned with a step size of 10nm and 1° to obtain a three-dimensional data table of area-depth-optical characteristics, with a total of 9×100×50=45,000 data points. Using MATLAB's curve fitting toolbox and based on the Gaussian process regression (GPR) algorithm, a mapping model of area-depth-optical characteristics to display effects is constructed. After 10-fold cross validation, the goodness of fit R² of the model reaches more than 0.98. On this basis, with display effect indicators such as parallax range, depth of field range, and interlacing effect as constraints, and grating width and reflection angle as optimization variables, a multi-objective constrained optimization model of grating parameters is constructed. The optimization model was solved by genetic algorithm (GA), with the population size set to 100, the crossover probability to 0.8, the mutation probability to 0.1, and the number of iterations to 500. The optimal grating parameter combination of 9 regions at 100 depth layers was obtained, and the parameter discretization accuracy reached 10nm and 1°. In the stereo display process, the 3D scene image was acquired at a speed of 30 frames / second by the binocular camera, and the disparity map was generated through the stereo matching algorithm SGM. The disparity search range was 0-100 pixels and the disparity resolution was 0.1 pixels. According to the disparity map and grating parameter combination, the parallax compensation mathematical model was used to resample the left and right eye views in the horizontal and vertical directions to compensate for the pixel displacement of the parallax in the range of 0.1-1 pixels, and the compensation accuracy reached the sub-pixel level. After subjective evaluation by 10 users, the depth of field range of 3D stereo display reached 800-1000mm, the parallax range was ±20 pixels, the interlacing effect was less than 5%, and the comprehensive score was 4.5 / 5 points.

[0118] The display effect of the predetermined layer information of each area is analyzed, the display performance of each area at different depth layers is measured and recorded, and the area suitable for displaying the first predetermined depth layer and the area suitable for displaying the second predetermined depth layer are analyzed and determined.

[0119] According to the resolution and size of the electronic ink screen, the screen is divided into several display areas by using the method of equally spaced grid division. The size and number of pixels of each area are the same, ensuring the uniformity and consistency of the area division, and providing a basis for the subsequent regional depth matching and parameter optimization. For each display area, according to its location and size, the corresponding first and second predetermined depth layer information is determined. Through optical simulation and experimental measurement, the optical characteristic parameters of the area at different depth layers, such as brightness, contrast, color saturation, etc., are obtained, and a three-dimensional mapping relationship between area-depth layer-optical characteristics is established. A high-speed camera is used to continuously shoot the display performance process of each area at different depth layers to obtain a dynamic image sequence with a time resolution of milliseconds. Then, image processing algorithms such as frame difference method and optical flow method are used to calculate the dynamic characteristic parameters such as the response time and grayscale change curve of the regional display state. At the same time, optical sensors such as spectrophotometers or colorimeters are used to accurately measure the static characteristic parameters such as brightness, contrast, and color of the display performance screen of each area at different depth layers. The spatial resolution reaches the millimeter level, the color resolution reaches 24-bit true color, and the measurement accuracy is better than 1%. The dynamic characteristic parameters and static characteristic parameters are associated with the depth layer to construct a three-dimensional data matrix of region-depth-display characteristics. Then, data mining algorithms such as association rule analysis and decision tree analysis are used to discover the distribution law and influencing factors of display characteristics at different depth layers, providing a decision basis for regional depth matching. The display characteristic parameters of each region at different depth layers are used as the depth attribute vector of the region. Then, clustering algorithms such as K-means and hierarchical clustering are used to cluster all regions into several regional groups with similar depth attributes. Each regional group corresponds to an optimal depth layer, that is, the display characteristics of all regions in the regional group at this depth layer are optimal. For each regional group, according to the depth value of its optimal depth layer, as well as the physical characteristics and optical parameters of the electronic ink screen, a mapping model of grating width, reflection angle and display characteristics is established, such as a polynomial regression model, a Gaussian process regression model, etc. Then, optimization algorithms such as genetic algorithms and particle swarm algorithms are used to search for the optimal grating parameter combination so that the overall display characteristics of the regional group are optimal. During the optimization process, the smooth transition and continuity of the grating parameters between adjacent area groups also need to be considered. Smooth regularization terms can be introduced into the fitness function, or the interpolation algorithm can be used to interpolate and smooth the grating parameters of adjacent area groups to avoid sudden changes and discontinuities in display quality at the area boundaries. During the 3D stereoscopic display process, the grating parameters and display content of each area are dynamically adjusted according to the changes in the depth map and the optimal matching scheme of the area-depth layer, and real-time switching and transition between depth layers are achieved to ensure the coherence and stability of the stereoscopic picture and improve the user's visual experience and immersion. Design a parallax adjustment mechanism and algorithm based on the parallax changes between different depth layers and the physiological characteristics of the human eye.According to the physiological parameters of the human eye, such as pupil distance, adjustment amplitude, etc., and the comfortable viewing distance range, a disparity-depth mapping model and a disparity-visual comfort model are established to determine the optimal disparity range and disparity distribution. For each frame of a 3D stereoscopic image, the disparity value of each area is calculated according to its depth map and the depth layer distribution of the display area to obtain the disparity distribution map of the entire screen. The disparity distribution map is compared with the optimal disparity range to find the area where the disparity exceeds the range and calculate its disparity compensation amount. The size of the compensation amount is proportional to the degree to which the disparity deviates from the optimal range, and the compensation direction is opposite to the deviation direction. According to the disparity compensation amount, the left and right eye views of each area are horizontally displaced and resampled to generate new left and right eye views after disparity adjustment. In order to ensure the continuity and consistency of objects before and after disparity adjustment, the disparity compensation amounts of adjacent areas need to be smoothed. Algorithms such as Gaussian filtering and bilateral filtering can be used to smooth the disparity compensation amount map to avoid problems such as parallax mutation and object breakage. While adjusting the parallax, it is also necessary to fine-tune the brightness and contrast of the left and right eye views to compensate for the problems such as inconsistent brightness and reduced contrast caused by the parallax adjustment, to ensure the visual quality and comfort of the final display image, and to maximize the perceived effect of the 3D stereoscopic display.

[0120] For example, for a 7-inch electronic ink screen with a resolution of 1600×1200, it is divided into a 4×3 grid to form 12 display areas, each containing 400×400 pixels. The optical properties of each area are simulated by Zemax Optic Studio software, and the depth layer is divided in steps of 0.1 within the relative depth range of 0.1 to 1.0, to obtain 10 predetermined depth layers. A high-speed camera is used to continuously shoot the display performance process of the electronic ink screen at different depth layers at a speed of 1000 frames per second, and an FPGA-based image processing module is used to calculate the response time of each area by the frame difference method, with the minimum detectable time difference of 0.1 milliseconds; at the same time, a spectral scanner with a resolution of 1nmx1nm and a measurement accuracy better than 0.5% is used to measure the brightness, contrast and color of the display performance screen of each area, and a three-dimensional data matrix of area-depth-display characteristics is obtained, containing 12x10x5=600 data points. The Apriori association rule mining algorithm was used to find the association rules between the depth layer, the region position and the display brightness and contrast, with the minimum support of 0.2 and the minimum confidence of 0.8 as the threshold. For example, when the depth layer is small and the region is located in the center of the screen, the display brightness and contrast are higher. Based on the association rules, the K-means clustering algorithm was used to cluster the 12 regions, with the goal of minimizing the inter-class variance and maximizing the intra-class similarity, and three region groups were obtained, which were suitable for displaying relative depth ranges of 0.1-0.3, 0.4-0.7 and 0.8-1.0, respectively. For each region group, a Gaussian process regression model of grating width, reflection angle and display brightness and contrast was established, and the particle swarm algorithm was used for parameter optimization, with the average relative error less than 5% as the optimization target. The search range was grating width 100-500nm, reflection angle 10°-60°, population size 50, evolutionary generation 100, convergence speed control factor 0.8, and inertia weight 0.5. In the optimization process, the cubic spline interpolation algorithm is used to interpolate and smooth the grating parameters between adjacent area groups to ensure the continuity of the grating parameter changes. In 3D stereoscopic display, based on the physiological parameters of human pupil distance of 65mm and adjustment amplitude of 2°, the optimal disparity range is calculated as ±1° through the disparity-depth mapping model, and the disparity is quantified with a step size of 0.1°. For each frame of the depth map, the disparity value of each area is calculated using the Census transformation algorithm to generate a disparity distribution map, which is compared with the optimal disparity range to obtain a disparity compensation map. The disparity compensation direction is opposite to the deviation direction, the compensation amplitude is proportional to the deviation, and the compensation coefficient is 0.8. When performing disparity compensation for the left and right eye views, the bilinear interpolation algorithm of 4×4 pixel blocks is used to resample the displaced pixels, and a 5×5 Gaussian smoothing filter is used to smooth the disparity compensation map to eliminate disparity mutations.Finally, a gamma correction algorithm is used to fine-tune the grayscale values ​​of the left and right eye views, with a gamma value of 2.2 to compensate for the 3% reduction in brightness and 5% reduction in contrast caused by parallax adjustment.

[0121] Based on the regional division plan and deep mapping basic data, a standard distribution of content types corresponding to each region is generated. Combined with real-time data and user feedback, the content distribution of each region is adjusted to determine the content type displayed in each region to achieve the desired effect.

[0122] According to the regional division scheme and the basic data of depth mapping, the depth features of each display area are analyzed, and feature vectors such as the depth distribution histogram and depth gradient histogram of the area are extracted. Combined with the predefined content type label system, large-scale regional display data, including the depth features of the area, content type labels, etc., are collected to build a training data set. Preprocess and feature engineering are performed on the deep features, such as normalization, feature selection, feature combination, etc., to extract the most discriminative feature subset. Select appropriate classification algorithms, such as logistic regression, support vector machine, random forest, etc., based on factors such as data scale, feature type, and computing resources. Use methods such as cross-validation and grid search to optimize the hyperparameters of the algorithm and improve classification performance. Evaluate the accuracy, precision, recall and other indicators of the classification model on the test set, compare the performance of different algorithms, select the optimal model, train the region-depth-content type classification model, and obtain the standard distribution of content types suitable for display in each area. Using data mining algorithms such as association rule mining and frequent item set mining, we can find the association rules and co-occurrence patterns between different regions, different depth layers, and different content types from massive historical data, and generate association rule libraries and frequent pattern libraries for regional content distribution as the basis for content distribution and recommendation. Collect multi-dimensional attribute data of users, such as demographic characteristics, behavioral characteristics, social characteristics, etc., and construct user feature vectors. Perform dimensionless processing on user characteristics, such as Min-Max normalization and Z-score normalization, to eliminate the dimensional influence between different attributes. Select appropriate user similarity measurement methods, such as Euclidean distance, cosine similarity, Pearson correlation coefficient, etc., according to data distribution and business needs. Use clustering algorithms such as K-means and DBSCAN to group users, and determine the optimal number of clusters and clustering results based on clustering effects and business interpretability. Perform statistical analysis on the content preferences of each user group, such as click-through rate, browsing time, etc., and construct a group-content type preference matrix for subsequent personalized recommendations. Build a user portrait model, divide users into multiple user groups, and model the content preferences of each user group. Collect user behavior data in real time, such as clicks, browsing, and favorites, and update user dynamic features. Use collaborative filtering algorithms, such as user-based collaborative filtering and item-based collaborative filtering, to generate a real-time recommendation list based on the user's historical behavior and the preferences of similar users. Use content-based recommendation algorithms, such as TF-IDF and Word2Vec, to recommend similar content based on the content features of the user's historical browsing. Use combined recommendation algorithms, such as weighted average and Stacking, to combine the results of multiple recommendation algorithms to generate a comprehensive recommendation list. Track and analyze user behavior data in real time, dynamically track and predict users' real-time interests, generate users' real-time content recommendation lists, and match them with regional content distribution standards to obtain real-time recommended content for each region.According to the user feedback data on the recommended content, such as click-through rate, dwell time, etc., the recommendation algorithm and regional content distribution standards are optimized and adjusted in real time. Reinforcement learning algorithms, such as Thompson Sampling, are used to balance exploration and utilization, dynamically adjust the recommendation strategy, and improve the overall revenue. Online learning algorithms, such as online gradient descent and online logistic regression, are used to update the parameters of the recommendation model in real time to adapt to changes in user preferences and maximize the overall user satisfaction and engagement. Online experimental methods such as A / B testing are used to randomly assign users to different experimental groups and control groups to expose different content distribution strategies. User feedback data on different strategies, such as click-through rate, conversion rate, dwell time, etc., are collected to construct the experimental result data set. Hypothesis testing methods, such as t-test and chi-square test, are used to evaluate whether the effect differences between different experimental groups are significant. Multi-armed bandit algorithms, such as UCB and Thompson Sampling, are used to dynamically allocate traffic, balance exploration and utilization, and maximize the overall revenue. Compare and test different regional content distribution schemes, evaluate the display effects of different schemes, and select the optimal content distribution scheme. Monitor the content display effect of each region in real time, collect user feedback through questionnaires, interviews and other methods, and evaluate the role of content in improving qualitative indicators such as user emotional resonance and reputation. In the process of data collection, fully consider user privacy protection, standardize data flow and access rights, and reduce the risk of data misuse. Based on the evaluation of display effect, continuously optimize and improve the content distribution strategy, such as adjusting algorithm parameters, improving feature engineering, introducing new data sources, etc., to continuously improve the pertinence and effectiveness of content display, and ultimately achieve the expected display goals.

[0123] For example, for an e-book reader with 500 display areas, the reading behavior data of 1 million users within 3 months are collected, including browsing time, number of clicks, collection ratio, etc., totaling 100 million records. The minimum-maximum normalization method is used to non-dimensionalize the behavioral indicators of different orders of magnitude, and then the improved cosine similarity algorithm is used to calculate the similarity matrix between users. The similarity threshold is set to 0.8, and the DBSCAN clustering algorithm is used to divide users into 50 groups. For each user group, the average browsing time, click-through rate and other indicators in different content categories are counted to construct a 50x10-dimensional group-content preference matrix. Using the association rule mining algorithm, with the minimum support of 0.05 and the minimum confidence of 0.3 as the threshold, 2000 association rules and 5000 frequent item sets between user groups, reading areas, and content categories are found. The logistic regression algorithm is used to classify the content types of the regions. With deep features and TF-IDF text features as input, 10 content types are multi-classified. Through 5-fold cross validation and grid search, the accuracy rate on the validation set is 85%, and the average AUC is above 0.9. In the online environment, user behavior data is updated every 30 seconds. A weighted serial recommendation architecture that integrates algorithms such as ItemCF, UserCF, and ContentBased is used to generate a personalized recommendation list of 10 in real time for each user. The online Thompson sampling algorithm is used for strategy adaptation to improve the click-through rate and user satisfaction of the recommendation. An AB test scheme with a traffic split ratio of 1:1:1:1 is used to compare 4 different personalized recommendation strategies, collect indicators such as user clicks, dwell time, and content completion rate. After two weeks of data accumulation, the paired t-test method is used to compare the significant differences between different strategies, and the best recommendation strategy is determined. Finally, a 28-question questionnaire is used to survey and evaluate 1,000 users, compare the improvement of content distribution effect before and after optimization, explore the rules, and continuously optimize and improve the intelligent distribution mechanism.

[0124] A genetic algorithm is used to output the optimal area combination solution based on the content type and its priority for grating width and reflection angle.

[0125] According to the display area division of the electronic ink screen and the content display requirements, different types of content are classified and labeled, and the corresponding relationship matrix between content type and display area is established. The grating width and reflection angle parameters of each content type are initialized as optimization variables of the genetic algorithm. For each display area, according to the type of content to be displayed, the value range and constraints of the grating width and reflection angle parameters are defined, and the objective function of parameter optimization is established. Considering the display quality and energy consumption indicators, the weighted summation method is used for quantification, and the weight coefficient is adjusted according to the actual needs and preferences. At the same time, a penalty function is set to punish solutions that violate the constraints and guide the algorithm to search for a reasonable and feasible solution space. Using real number coding, the grating width and reflection angle parameters of each display area are mapped to the individual genes of the genetic algorithm. The initial population is generated by random sampling and other methods, and the fitness function value of each individual is calculated according to the parameter value of the individual gene, and the population is ranked. The individuals in the population are subjected to genetic operations such as selection, crossover and mutation. The selection operation adopts roulette or tournament methods to give priority to individuals with high fitness for reproduction. The crossover operation can select single-point crossover or multi-point crossover to exchange the gene fragments of two parent individuals at one or more randomly selected gene positions to generate new offspring individuals. The mutation operation can select uniform mutation or non-uniform mutation to randomly perturb the value on each gene position with the same or different probability. The perturbation probability and amplitude of non-uniform mutation decrease with the increase of evolutionary generations, which is conducive to local fine-tuning. On the basis of genetic operations, the individual genes are corrected and adjusted according to the priority of the content type, and the priority weight matrix is ​​set to assign different weight coefficients to different content types. When evaluating the fitness, it is multiplied with the display quality and energy consumption indicators to obtain the weighted fitness value, so as to achieve the preference for high-priority content types. For the gene positions corresponding to the high-priority content types, their crossover and mutation probabilities are reduced to maintain excellent characteristics; for the gene positions corresponding to the low-priority content types, their crossover and mutation probabilities are increased to accelerate elimination and updating. For individuals that violate the priority constraint, a penalty term is imposed on the fitness value to reduce the probability of their selection. The fitness function value of the new generation population after the genetic operation is recalculated, and the elites are retained and the population is merged with the previous generation population to control the population size to be stable within a certain range. At the same time, the diversity indicators of the population, such as gene entropy and clustering degree, are monitored. When the diversity indicators are lower than the set threshold, the population reset and reinitialization operation is triggered. A few individuals with the highest fitness can be retained, and the remaining individuals are randomly generated, or the current individuals are highly mutated, or new random individuals are introduced to merge with the current population to improve population diversity and avoid premature convergence.The above genetic algorithm steps are iterated until the preset convergence conditions are met, such as no significant improvement in the fitness function value for consecutive generations, or the number of iterations reaches the upper limit, etc. The individual with the highest fitness in the current population is output as the optimal display area combination scheme and grating parameter configuration to guide the display optimization and content distribution of the electronic ink screen. During the algorithm execution process, the convergence speed and optimization effect of the algorithm are evaluated by real-time monitoring and recording of indicators such as the optimal individual and average fitness of each generation of the population, and the algorithm parameters are dynamically adjusted as needed, such as appropriately adjusting the crossover probability, mutation probability, penalty coefficient, etc., to balance local search and global exploration, and improve the convergence speed of the algorithm and the quality of the solution.

[0126] For example, for an 8-inch electronic ink screen, it is divided into 16 display areas, and the resolution of each area is 200×200 pixels. According to different content types, such as text, image, video, etc., fuzzy clustering algorithm is used to classify them, and the AHP hierarchy analysis method is used to calculate the priority weights of each type, where the weight of text type is 0.5, the weight of image type is 0.3, and the weight of video type is 0.2. For each display area, the grating width ranges from 200-400nm, and the reflection angle ranges from 10° to 60°. The combination of grating width and reflection angle is used as the optimization variable, and the weighted normalization method is used to quantify indicators such as display contrast and energy consumption as optimization targets, and a penalty function is set. For solutions with a grating width less than 200nm or greater than 400nm, and solutions with a reflection angle less than 10° or greater than 60°, a 10-fold penalty term is applied to the objective function value. Real number coding is used to randomly generate 50 initial solutions, each of which consists of 32 variables. The first 16 variables represent the grating width, and the last 16 variables represent the reflection angle. In genetic operations, the selection operation adopts tournament selection. Five individuals are randomly selected from the population each time, and the two with the worst objective function values ​​are eliminated, and the best two are retained for crossover. The crossover operation adopts arithmetic crossover, and the weighted average of the variables of the two parent individuals is taken, with weights of 0.3 and 0.7 respectively, to generate new offspring individuals. The mutation operation adopts non-uniform mutation, and the initial value of the mutation probability is 0.5, which decays to 0.95 times the original value every 10 generations. The initial value of the mutation amplitude is 20% of the variable value range, which decays to 0.9 times the original value every 10 generations. The display areas are sorted according to the priority weight of the content type. For the top 30% of the areas, the crossover probability of their variables is reduced by 20%, and the mutation probability is reduced by 30%; for the bottom 30% of the areas, the crossover probability of their variables is increased by 20%, and the mutation probability is increased by 30%. The population size of each generation is kept at 50, and the elite retention strategy selects the two individuals with the highest fitness values ​​to enter the next generation directly. The population diversity evaluation index uses the gene position entropy. When the average gene position entropy of the population for five consecutive generations is lower than 2.5, the population reset is triggered, and the five individuals with the highest fitness values ​​are randomly selected and retained, and the remaining individuals are all randomly regenerated. The algorithm iterates for 500 generations, and the optimal objective function value and average objective function value of each generation are recorded. Finally, the display area raster parameter combination corresponding to the individual with the highest objective function value is output as the optimal solution to the problem.

[0127] According to the requirements of smoothness and clarity of the displayed content, the corresponding grating width and reflection angle are matched. Through the genetic algorithm, the dynamic video content is prioritized in the area where the grating width and reflection angle are balanced, and the static graphic content is placed in the area with narrow grating and predetermined reflection angle to highlight its details and contrast.

[0128] According to the display area division of the electronic ink screen, different types of display content are identified and classified. Image classification algorithms based on deep learning, such as convolutional neural network (CNN) and residual network (ResNet), are used to extract features and judge the type of input display content, and obtain different types of content labels such as dynamic video, static chart, text, etc., which are used for subsequent regional matching and optimization. For each type of display content, a mapping relationship model between content type and grating width and reflection angle is established according to its fluency and clarity requirements. A large-scale electronic ink screen display content data set is constructed, including display samples of different types and different parameter configurations, as well as corresponding subjective and objective evaluation indicators such as fluency and clarity, as training data sets. For the fluency index, objective indicators such as frame rate, frame delay, frame loss and subjective indicators such as user subjective rating are used to quantify it into a score of 0-100 or a level of 1-5; for the clarity index, objective indicators such as contrast, sharpness, color saturation and subjective indicators such as user subjective rating are used to quantify it into a score of 0-100 or a level of 1-5. Using deep learning algorithms, such as convolutional neural networks (CNNs) and long short-term memory networks (LSTMs), the image features of the displayed content and the grating parameter configuration are used as input, and the fluency and clarity indicators are used as output to train an end-to-end mapping model. During model training, regularization techniques such as cross-validation and early stopping are used to prevent overfitting; hyperparameter optimization techniques such as grid search and Bayesian optimization are used to automatically adjust the model's hyperparameters to find the optimal model structure and parameter configuration. Newly collected user feedback data is regularly used to incrementally train and fine-tune the mapping model to achieve continuous optimization and adaptive capabilities of the model. For dynamic video content, by analyzing its frame rate, motion speed, scene complexity and other features, the appropriate grating width and reflection angle parameter range are matched to ensure the smoothness and continuity of video playback. The genetic algorithm is used to optimize the parameter combination, taking the average frame rate as the primary optimization target, while taking into account indicators such as the peak signal-to-noise ratio between frames, designing a weighted sum fitness function, searching for the optimal grating parameter combination, and giving priority to allocating it to the display area with balanced grating width and reflection angle, avoiding problems such as dynamic video freeze and afterimage. For static chart content, by analyzing its edge, texture, color and other features, matching the appropriate grating width and reflection angle parameter range to ensure the clarity and readability of the chart display. The genetic algorithm is used to optimize the parameter combination, taking contrast as the primary optimization target, while taking into account indicators such as edge sharpness, designing a weighted sum fitness function, searching for the optimal grating parameter combination, and giving priority to allocating it to the display area with narrow grating and predetermined reflection angle, highlighting the details and contrast of the chart, and improving the effect of information transmission.In the optimization process of the genetic algorithm, the real number coding method is used to map the grating width and reflection angle parameters to genotypes. Through selection, crossover, mutation and other operations, new parameter combinations are generated, and their optimization performance is evaluated according to the fitness function. The fitness function integrates the fluency index of dynamic video content and the clarity index of static chart content, designs a multi-objective fitness function, performs weighted summation or non-dominated sorting on different indicators, balances the weights of fluency and clarity, and obtains a comprehensive display quality evaluation. A penalty term is introduced in the fitness function. For solutions that exceed the grating width and reflection angle constraint range, a penalty function proportional to the degree of constraint violation is applied to reduce its fitness value, and guide the algorithm to search for feasible solutions that meet the constraints. In the display system of the electronic ink screen, a user feedback interface and evaluation module are embedded to allow users to subjectively score or give feedback on the fluency and clarity of the current display content. At the same time, through sensors or image analysis algorithms, objectively measure indicators such as the frame rate and contrast of the display content, and collect user feedback data and objective evaluation data in real time. The collected data is automatically uploaded to the cloud data analysis platform in a certain format and frequency, and associated with metadata such as display content information and grating parameter configuration to form new training data samples. A machine learning pipeline for incremental learning and online optimization is built on the cloud, and tasks such as data preprocessing, feature engineering, model training, and model evaluation are regularly run. The mapping model between content type and grating parameters is incrementally trained and fine-tuned using newly collected data samples to continuously improve the accuracy and generalization performance of the model. The optimized mapping model is sent to the display system of the electronic ink screen in the form of an incremental update package to replace the original model file, and realize online update and adaptive optimization of the display strategy. Online experimental platforms such as A / B testing and multi-armed bandit are built on the cloud to conduct random diversion tests on different versions of mapping models and display strategies. Through user feedback and statistical analysis of objective indicators, the display effects and user experience of different strategies are evaluated, and the optimal strategy is selected for full deployment and promotion. The multi-armed bandit algorithm is used to dynamically adjust the diversion ratio of different strategies, balance exploration and utilization, and continuously optimize the display strategy. For other types of display content (such as text), heuristic rules or constraint satisfaction-based search algorithms are used to allocate them to the remaining display area, and the grating width and reflection angle parameters are fine-tuned according to their smoothness and clarity requirements to ensure that the display effect of the entire electronic ink screen is optimal.

[0129] For example, a 10.3-inch electronic ink screen is optimized for content adaptive display. First, the pre-trained ResNet-50 model is used to classify the display content, and the classification accuracy on the ImageNet dataset reaches more than 92%. A special display quality dataset is constructed for different types of content. 10,000 samples are extracted for each type. Each sample contains 100 different display images under grating width (100-300nm) and reflection angle (10-60°) combinations, as well as corresponding frame rate, contrast and other indicators and user ratings. The total data volume exceeds 10 million. Using the CNN+LSTM combined model, the image and parameter features are input into CNN to extract spatial features, and then into LSTM to learn temporal association. After 10 epochs of training, the mean square error of frame rate and contrast prediction on the test set is reduced to less than 2% and 3%, respectively. The model was deployed in the display control system of the electronic ink screen. For video content, the frame rate was the main optimization target, and the weights of contrast and peak signal-to-noise ratio were set to 0.6 and 0.3 respectively. The genetic algorithm was used to search for the optimal parameter combination, with a population size of 50, 100 iterations, a crossover probability of 0.8, and a mutation probability of 0.1. The average frame rate was increased by 20% and the peak signal-to-noise ratio was increased by 1.5dB. For chart content, the contrast was the main optimization target, the weight of sharpness was set to 0.7, and the genetic algorithm was used to search for the optimal parameter combination, with a population size of 30, 50 iterations, a crossover probability of 0.7, and a mutation probability of 0.15. The contrast was increased by 15% and the sharpness was increased by 10%. A penalty term was added to the fitness function, and the fitness value was halved for solutions with grating width exceeding 250nm or reflection angle exceeding 50°. During the operation of the system, 5% of the user feedback scores on display quality were randomly collected every day, and the average value was taken to update the display quality data set. The mapping model was incrementally trained with the newly added data every week, and the learning rate was set to 1 / 10 of the original. The average user rating increased from 4.1 to 4.5 (out of 5) for four consecutive weeks, indicating that the display optimization strategy can continuously improve the user experience.

[0130] By utilizing the area division scheme and depth mapping basic data, combined with user feedback and real-time monitoring data, a random forest model and fuzzy logic control system are used to establish a correlation model between content type and electronic ink screen display effect. By real-time monitoring of the actual display results on the electronic ink screen and comparing and analyzing them with the expected results, the content is adjusted or the grating parameters are modified according to the grating width and reflection angle of each display area.

[0131] According to the physical characteristics and display requirements of the electronic ink screen, an adaptive segmentation algorithm based on GraphCut is used to divide the screen into multiple rectangular display areas. First, the semantic segmentation information of the display content is introduced, and the display image is semantically annotated at the pixel level through a deep learning model (such as FCN, UNet, etc.), and different content types (such as text, images, videos, etc.) and display objects (such as titles, text, icons, etc.) are identified to generate a semantic segmentation map. Then, the semantic segmentation map and pixel density information are combined to merge regions with similar semantics and similar pixel density into one display area, while considering the shape regularity and size balance of the region to avoid the appearance of too slender or too narrow areas. Finally, the divided display areas are encoded and annotated to generate a region division map, and the attribute information such as the position, size, pixel density, and semantic category of each region is extracted for subsequent parameter optimization and dynamic adjustment. Using a binocular camera or a structured light sensor, a multi-view image or depth map of the electronic ink screen display content is obtained as the input data for stereo vision reconstruction. Using learning-based stereo matching algorithms, such as CNN-basedStereoMatching and GC-Net, the convolutional neural network is used to extract and match the features of the left and right view images to generate dense disparity maps or depth maps. The generated depth map is post-processed and optimized, such as median filtering and bilateral filtering, to eliminate noise and outliers and improve the smoothness and continuity of the depth layer. According to the depth map and display area information, the pixel distribution, edge contour, texture features, etc. of each area at different depth layers are extracted to generate structured depth mapping data to provide data support for subsequent content type division and parameter optimization. A user feedback collection mechanism is established to regularly send display effect evaluation questionnaires to users to collect users' subjective ratings and opinions on different content types in different display areas. At the same time, the user's display preferences, usage environment and other contextual information are recorded to build a user feedback database to provide a reference for the formulation of display optimization strategies. An optical sensor array and an image processing unit are integrated on the electronic ink screen to collect screen display images and ambient lighting conditions in real time, extract objective indicators such as display brightness, contrast, and color saturation, generate real-time monitoring data streams, associate with depth mapping data and user feedback data, and build a multi-dimensional display effect evaluation data set. An integrated learning strategy is used to train multiple random forest models, such as Bagging and RandomSubspace. By sampling and feature sampling the display effect evaluation data set, multiple different training subsets and feature subsets are generated, and multiple decision trees are trained separately. The input features are normalized, discretized, combined, and other transformations are performed to extract high-dimensional features such as color histograms, gradient histograms, and frequency domain features of the display content, and feature selection algorithms (such as tree-based feature importance evaluation) are used to select the most relevant feature subsets.When training each decision tree, the Gini index or information gain ratio is used as the basis for node division. The optimal division features and thresholds are automatically selected through heuristic methods (such as optimal division point search) to generate leaf nodes that are as pure as possible. At the same time, early stopping strategy and pruning strategy are introduced to control the growth depth and complexity of the tree to avoid overfitting. The trained random forest model is evaluated and optimized. Cross-validation, grid search and other methods are used to select the optimal hyperparameter combination (such as the number of trees, the maximum depth of the tree, the minimum number of samples of the node, etc.) to improve the prediction accuracy and stability of the model, and obtain a nonlinear association model between content type and display effect to guide the generation of display optimization strategy. For different content types and display areas, combined with depth mapping data and association models, a fuzzy logic control system is used to design a set of rule-based parameter optimization strategies. According to the display effect evaluation index and the visual perception characteristics of the human eye, a fuzzy rule base is designed to map the index value to fuzzy language variables (such as "low", "medium", "high", etc.), and the membership function of the fuzzy set (such as triangle, trapezoid, Gaussian function, etc.) is defined to establish a fuzzy evaluation system for display effect. The grating width and reflection angle parameters are fuzzified, and the fuzzy subsets of the parameters (such as "narrow", "moderate", "wide", etc.) and the membership function are defined to establish the fuzzy adjustment space of the parameters. Based on expert knowledge and experience, fuzzy control rules are designed to establish a correlation between display effect evaluation and parameter adjustment, such as "if the contrast is low and the clarity is high, the grating width is moderate and the reflection angle is small", etc., to form a complete fuzzy rule base. Fuzzy reasoning methods such as Mamdani reasoning or Takagi-Sugeno reasoning are used to infer the fuzzy language values ​​of parameter adjustment based on the fuzzy language values ​​of display effect evaluation through fuzzy matching, rule activation and synthesis. The fuzzy parameter values ​​obtained by reasoning are defuzzified, such as the centroid method, the maximum membership method, etc., to convert the fuzzy language values ​​into accurate parameter values, generate a continuous and smooth parameter adjustment curve, and realize flexible optimization of display effects. During the display process of the electronic ink screen, the actual display results on the screen are monitored in real time, key frame images are extracted, and subjective and objective evaluation indicators such as contrast, clarity, and color saturation are calculated to form a multi-dimensional indicator vector. Compare the indicator vector of the actual display result with the indicator vector of the expected result given by the association model, and calculate the distance measurement between them, such as Euclidean distance, Mahalanobis distance, etc., as a quantitative indicator of display deviation. Perform statistical analysis on historical display data, calculate the distribution characteristics of display deviation, such as mean, variance, quantile, etc., and automatically determine the threshold of display deviation based on empirical rules (such as 3-sigma principle) or statistical hypothesis tests (such as t test, F test).When the display deviation calculated in real time exceeds the threshold, the parameter optimization strategy is triggered. According to the size and direction of the deviation, proportional-integral-differential (PID) control or adaptive control methods are used to dynamically adjust the grating width and reflection angle parameters, where Kp and Ki are proportional coefficients and integral coefficients, and Integral is the cumulative value of the deviation. In the process of parameter adjustment, set a reasonable adjustment step and boundary conditions to avoid excessive parameter fluctuations or exceeding the feasible domain. At the same time, consider the mutual influence and constraint relationship between different regions, coordinate the parameter adjustment of multiple regions, and realize dynamic closed-loop optimization of display effects.

[0132] For example, for a 13.3-inch electronic ink screen, a semantic segmentation model based on UNet is used for pre-training on the PASCAL VOC dataset, with an average IoU of more than 85%. Then, it is fine-tuned on a dataset containing 500 electronic ink screen display images to obtain pixel-level display content semantic annotation results. The semantic segmentation results are optimized by graph cutting using the Grabcut algorithm of OpenCV, and the screen is divided into 50 display areas, with an average of 10,000 pixels per area, combined with the pixel density clustering algorithm DBSCAN. The depth map of the display content is obtained through the binocular structured light sensor Intel Real Sense D435, with a resolution of 1280x720 and a depth error of less than 2%. The stereo matching algorithm PSMNet is used to generate a dense disparity map, and a smooth and continuous depth map is obtained after 5x5 median filtering and guided filtering. The depth map is mapped to the display area, and the histogram distribution and gradient histogram distribution of each area in the depth range of 0-255 are statistically analyzed. The subjective ratings of 100 users on the display effects of different contents in different areas were collected through the online questionnaire platform SurveyMonkey, using a 5-point Likert scale, and recording the user's age, gender, occupation and other user portrait data. The sensor array was used to collect parameters such as ambient light intensity (0-100000 Lux), display brightness (0-500 nit), color temperature (2000-7000 K) in real time, and the display effect evaluation data set was constructed together with the subjective rating data, with a total of 50x100=5000 samples. Feature engineering was performed on the data set to extract color histograms (16x16x16 bins), LBP texture features (256 dims), FFT frequency domain features (32x32 dims), etc., and the GBDT feature selection algorithm was used to select the 50 most important features, train 30 decision trees with a maximum depth of 10, generate a random forest model, and through 10-fold cross validation, the average accuracy rate reached more than 90%. Based on the fuzzy evaluation of display effect, a rule base containing 25 fuzzy rules is designed. The input variables are indicators such as contrast, clarity, and color saturation, and the output variables are grating width and reflection angle parameters. The Mamdani reasoning method is used to convert fuzzy rules into accurate parameter adjustment through maximum-minimum synthesis and center of gravity defuzzification. The display deviation is calculated in real time, for example, the color difference DeltaE between the actual display image and the expected image is calculated in the sRGB color space, and the threshold is set to 10. When the deviation exceeds the threshold, the PID controller is triggered to dynamically adjust the grating width (100-300nm) and reflection angle (10°-60°) according to the size of the deviation, and the adjustment formula is adjusted.

[0133] In the description of the present application, it should be understood that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present application and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present application.

[0134] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of this application.

Claims

1. A 3D display method for an electronic ink screen, characterized in that: The following steps are involved: Analyze the impact: Analyze the grating width and reflection angle of the electronic ink screen, including measuring and recording the light reflection characteristics under different grating widths, and the impact of the reflection angle on image clarity and contrast, to obtain the impact of different grating widths and reflection angles on screen display; Area division: According to the effects of different grating widths and reflection angles on screen display, the screen is divided into multiple display areas, classified according to grating width and reflection angle, and each area is matched with the corresponding grating width and reflection angle values ​​according to the aspect ratio of the screen to obtain an area division plan optimized for the screen; Establishing mapping relationship: According to the area division scheme, determine the mapping relationship between each display area and the 3D image information, obtain the display effect of each area at different depth layers according to the mapping relationship, obtain the expected display parameters, determine the stereoscopic display effect of the screen, and optimize the stereoscopic depth of the image to obtain the depth mapping basic data; Adjust content distribution: Generate a standard distribution of content types for each region based on the regional division plan and deep mapping basic data. Combine real-time data and user feedback to adjust the content distribution of each region so that the content types displayed in each region achieve the expected effect. Output combination scheme: Genetic algorithm is used to output the optimal regional combination scheme according to the content type and its priority for grating width and reflection angle; Adjust display content or parameters: Utilize the area division scheme and depth mapping basic data, combined with user feedback and real-time monitoring data, adopt the random forest model and fuzzy logic control system, establish the correlation model between content type and electronic ink screen display effect, monitor the actual display results on the electronic ink screen in real time, and compare and analyze with the expected results, adjust the content or modify the grating parameters according to the grating width and reflection angle of each display area.

2. The electronic ink screen 3D stereoscopic display method according to claim 1, characterized in that: The steps of analyzing the impact effect are specifically as follows: According to different grating widths and materials, the reflection angle, absorptivity, transmittance and refractive index of light on the grating surface are measured and recorded to obtain the light reflection characteristic data under different grating widths and material combinations; Analyze the reflection angle distribution under different grating parameters, use light under different reflection angles to illuminate the electronic ink screen, collect the screen display image through the image sensor, and quantitatively evaluate the image clarity and contrast to obtain the effect data of different reflection angles on the image quality of the screen display; A mathematical model is established based on the data between grating parameters, reflection angle and image quality. The grating width, material and reflection angle are used as input features and the image quality evaluation index is used as the output target. A regression model is trained to analyze the effects of different grating widths and reflection angles on screen display.

3. The electronic ink screen 3D display method according to claim 1, characterized in that: The area division steps are specifically as follows: According to the effects of different grating widths and reflection angles on the display effect of the electronic ink screen, experimental data are obtained through optical testing, including the reflected light intensity, reflection angle distribution, contrast, and color saturation under different grating parameter combinations, and a data set between grating parameters and display effects is constructed; Data preprocessing technology is used to clean, normalize and select features of the acquired optical characteristic data, remove noise data and redundant features, evaluate the importance of features through filtering or wrapping methods, select key features that best reflect the relationship between grating parameters and display effects, and construct an optimized data set; The optimized data set is input into the K-means clustering algorithm. By calculating the clustering performance index under different K values, the K value with the best performance and the appropriate number of clusters is selected in combination with the elbow rule or the Silhouette rule. Cluster analysis is performed to divide the display effect data points under different grating parameter combinations into K cluster centers. Each cluster represents a typical display effect mode, and the display area division scheme of the electronic ink screen is obtained. For each display area, the optimal grating width and reflection angle parameter range corresponding to the area is determined according to the cluster center to which it belongs, and a nonlinear mapping relationship between the grating parameters and the display effect in the area is established to obtain the grating parameter optimization strategy for each display area; Taking into account the aspect ratio and resolution of the electronic ink screen, the screen is divided into several grid units, each of which corresponds to a display area. According to the position and size of each grid unit, the corresponding grating width and reflection angle parameters are calculated to generate a regional parameter mapping table; According to the mapping table, the optimal grating parameter combination is assigned to each grid unit, and additional parameter adjustment and optimization are performed on the grid units in the edge area of ​​the screen to generate a regional layout map that is compatible with the screen specifications; Furthermore, according to the effects of different grating widths and reflection angles on screen display, each area is matched with an image with predetermined contrast content or predetermined detail richness. The image with predetermined contrast content or predetermined detail richness will be allocated to a predetermined display area, and other content will be allocated to other areas.

4. The electronic ink screen 3D display method according to claim 3, characterized in that: According to the effects of different grating widths and reflection angles on screen display, each area is matched with images with predetermined contrast content or predetermined detail richness. Images with predetermined contrast content or predetermined detail richness will be allocated to the predetermined display area, and other content will be allocated to other areas. Specifically: According to the display area division scheme of the screen, the grating width and reflection angle of each area are analyzed, and the mapping relationship between regional grating parameters and display effects is established. Through optical simulation and measured data, the display effect evaluation index of different areas under different grating parameters is obtained; According to the usage scenario and user needs, key display areas are pre-set as the scheduled display areas; An image processing algorithm is used to analyze the image content to be displayed, extract the image contrast information and detail richness information, and obtain the display effect evaluation index of the image content; For contrast, the global contrast and local contrast of the image are calculated, measured by the grayscale standard deviation and gradient amplitude mean of the image, respectively. According to the preset global and local contrast thresholds, the image is divided into high contrast and low contrast categories; For detail richness, the edge density and texture complexity of the image are calculated, and the edge pixel ratio obtained by the Canny edge detection algorithm and the entropy of the texture histogram obtained by the local binary pattern operator are used to measure them respectively. According to the preset edge density and texture complexity thresholds, the image is divided into two categories: high detail and low detail, and the classification labels of the image content are obtained. The number and proportion of each category of image content are counted; Matching the classification label of the image content with the display effect evaluation index of the display area, and finding the optimal image content, including a matching scheme of a predetermined contrast content or a predetermined detail richness image and the display area through the Hungarian algorithm or the KM algorithm; During the matching process, a higher priority is assigned to the predetermined display area, and image contents with high contrast and high details are preferentially assigned to the predetermined display area; For non-predetermined display areas, corresponding image content is allocated according to the quality of their display effects; According to the matching results between the image content and the display area, a multivariate nonlinear regression model is constructed, with the grating width and the reflection angle as independent variables and the display effect evaluation index as the dependent variable, and the least square method or gradient descent method is used to fit the model parameters; An optimization objective function is constructed for each region. The objective function is composed of the difference between the display effect evaluation index predicted by the regression model and the matched image content characteristics and the regularization term of the grating parameters. The optimal grating parameter combination of each region is solved by the gradient descent method or the quasi-Newton method so that the display effect of each region is adapted to the characteristics of the matched image content. According to the extracted image content features, the classification labels of the image content are updated in real time, and the region matching algorithm and grating parameter optimization algorithm are executed. An incremental update strategy is adopted to reclassify and match only the image content that has changed significantly.

5. The electronic ink screen 3D stereoscopic display method according to claim 1, characterized in that: The steps of establishing the mapping relationship are specifically as follows: According to the physical characteristics and display requirements of the electronic ink screen, the screen is divided into multiple display areas using a layered division method, each area has a different grating width and reflection angle; For the input 3D image, a stereo matching algorithm is used to match the corresponding points of the left and right views, calculate the disparity value of the pixel points, and then triangulate the depth value based on the disparity and camera parameters to obtain the depth map; According to the display area of ​​the screen, the depth map is divided into blocks and quantized to obtain the pixel distribution of each area at different depth layers; Based on the display area division and depth layer division, the optical characteristics of each area at different depth layers are modeled to establish a three-dimensional data table of area-depth-optical characteristics; According to the expected effects of 3D stereoscopic display, including depth of field range, parallax size and interlacing effect, the optical characteristics of each area at different depth layers are quantitatively analyzed and evaluated to obtain a set of objective indicators for measuring the display effect; The area-depth-optical characteristic data table is used as a training sample, and the display effect evaluation index is used as a training target. A three-dimensional mapping model is obtained through training to predict the display effect of each area at different depth layers. According to the importance and weight of the display effect index, the three-dimensional mapping model is optimized and simplified to obtain a simplified mapping relationship expression or mapping function for real-time calculation of the expected display parameters of each area at different depth layers, wherein the expected display parameters include grating width and reflection angle; The grating parameter optimization problem is modeled as a multi-objective constrained optimization problem, with the expected display parameters as the optimization target and the mapping relationship between the display area and the depth layer as the constraint condition, and the objective function and constraint equation of the optimization model are established; According to the characteristics of the objective function and constraint equations, a suitable optimization algorithm is selected, and according to the characteristics and parameter settings of the selected optimization algorithm, the optimization model is solved to obtain the global optimal or local optimal grating parameter combination; The optimized grating parameter combination is simulated and tested in real objects to evaluate its actual display effect at different depth layers until the actual display effect meets the display effect requirements; Furthermore, the display effect of the predetermined layer information of each area is analyzed to obtain the display performance of each area at different depth layers, and the area suitable for displaying the first predetermined depth layer and the area suitable for displaying the second predetermined depth layer are analyzed and determined.

6. The electronic ink screen 3D display method according to claim 5, characterized in that: The display effect of the predetermined layer information of each area is analyzed to obtain the display performance of each area at different depth layers, and the area suitable for displaying the first predetermined depth layer and the area suitable for displaying the second predetermined depth layer are analyzed and determined to be: According to the resolution and size of the electronic ink screen, the screen is divided into several display areas by using an equidistant grid division method, and the size and number of pixels of each display area are the same; For each display area, determine the first predetermined depth layer information and the second predetermined depth layer information corresponding to the area according to its location and size, obtain the optical characteristic parameters of the area at different depth layers through optical simulation and experimental measurement, including brightness, contrast and color saturation, and establish a three-dimensional mapping relationship between area-depth layer-optical characteristic; Continuously acquire images of the display performance process of each display area at different depth layers to obtain a dynamic image sequence, and then use an image processing algorithm to calculate the dynamic characteristic parameters of each display area, wherein the dynamic characteristic parameters include the response time of the display state and the grayscale change curve; At the same time, an optical sensor is used to measure and obtain the optical characteristic parameters of the display performance picture of each display area at different depth layers, which are recorded as static characteristic parameters; The dynamic characteristic parameters and the static characteristic parameters are associated with the depth layer to construct a three-dimensional data matrix of region-depth-display characteristics. Based on the obtained three-dimensional data matrix, the distribution law and influencing factors of the display characteristics at different depth layers are obtained; The display characteristic parameters of each region at different depth layers are taken as the depth attribute vector of the region, and then a clustering algorithm is used to cluster all regions into several region groups with similar depth attributes. Each region group corresponds to an optimal depth layer, that is, the display characteristics of all regions in the region group at this depth layer are optimal.

7. The electronic ink screen 3D display method according to claim 1, characterized in that: The step of adjusting content distribution comprises: According to the area division scheme and the depth mapping basic data, the depth features of each display area are analyzed, and the depth distribution histogram and depth gradient histogram of the area are extracted as feature vectors. In combination with the predefined content type label system, the regional display data, including the depth features and content type labels of the area, are collected to construct a training data set; Preprocess and feature engineer deep features to extract the most discriminative feature subset; Select a classification algorithm based on data size, feature type, and computing resources; Cross-validation and grid search methods are used to optimize the algorithm's hyperparameters to improve classification performance; Evaluate the accuracy, precision, and recall of the classification model on the test set, compare the performance of different algorithms, select the optimal model, train the region-depth-content type classification model, and obtain the standard distribution of content types suitable for display in each region; Using data mining algorithms, we can discover the association rules and co-occurrence patterns between different regions, different depth layers, and different content types from historical data, and generate association rule libraries and frequent pattern libraries for regional content distribution as the basis for content distribution and recommendation. Collect multi-dimensional attribute data of users and construct user feature vectors; Non-dimensionalize user features and select a user similarity measurement method; Use clustering algorithms to group users, and determine the optimal number of clusters and clustering results based on clustering effects and business interpretability; Conduct statistical analysis on the content preferences of each user group and construct a group-content type preference matrix; Build a user portrait model, divide users into multiple user groups, and model the content preferences of each user group; Collect user behavior data in real time and recommend content based on the behavior data; Conduct comparative tests on different regional content distribution plans, evaluate the display effects of different plans, and select the optimal content distribution plan.

8. The electronic ink screen 3D stereoscopic display method according to claim 1, characterized in that: The steps of outputting the combination scheme are specifically as follows: According to the display area division of the electronic ink screen and the content display requirements, different types of content are classified and labeled, a correspondence matrix between content type and display area is established, and the grating width and reflection angle parameters of each content type are initialized as optimization variables of the genetic algorithm; For each display area, according to the type of content to be displayed, the value range and constraints of the grating width and reflection angle parameters are defined, and the objective function of parameter optimization is established. The indicators of display quality and energy consumption are comprehensively considered and quantified by weighted summation. The weight coefficient is adjusted according to actual needs and preferences; Using real number coding, the grating width and reflection angle parameters of each display area are mapped to the individual genes of the genetic algorithm. The initial population is generated by random sampling, and the fitness function value of each individual is calculated according to the parameter value of the individual gene, and the population is sorted. Perform genetic manipulation on individuals in the population, giving priority to individuals with high fitness for reproduction; According to the priority of the content type, the individual genes are modified and adjusted, and a priority weight matrix is ​​set to assign different weight coefficients to different content types. When evaluating fitness, the weight coefficients are multiplied with the display quality and energy consumption indicators to obtain the weighted fitness value, thereby realizing the preference for high-priority content types. For gene positions corresponding to high-priority content types, reduce their crossover and mutation probabilities to maintain excellent characteristics; For gene positions corresponding to low-priority content types, their crossover and mutation probabilities are increased to speed up elimination and updating; For individuals that violate the priority constraints, a penalty term is imposed on the fitness value to reduce their probability of being selected; Recalculate the fitness function value of the new generation population after genetic operation, and retain the elite and merge the population with the previous generation population to control the population size to be stable within a certain range; In addition, according to the requirements of smoothness and clarity of the displayed content, the corresponding grating width and reflection angle are matched. Through a genetic algorithm, dynamic video content is prioritized in areas where the grating width and reflection angle are balanced, and static graphic content is placed in areas where the grating is narrow and the reflection angle is predetermined.

9. The electronic ink screen 3D stereoscopic display method according to claim 8, characterized in that: According to the requirements of the smoothness and clarity of the displayed content, the corresponding grating width and reflection angle are matched. Through the genetic algorithm, the dynamic video content is preferentially arranged in the area with balanced grating width and reflection angle, and the static chart content is placed in the area with narrow grating and predetermined reflection angle. Specifically: According to the display area division of the electronic ink screen, different types of display content are identified and classified. The image classification algorithm based on deep learning is used to extract features and judge the type of input display content, and obtain different types of content labels such as dynamic videos, static charts, and texts for subsequent area matching and optimization. For each type of display content, a mapping relationship model between the content type and the grating width and reflection angle is established according to the requirements of its fluency and clarity; Construct an electronic ink screen display content dataset, including display samples of different types and parameter configurations, as well as corresponding evaluation indicators of fluency and clarity, as a training dataset; For the fluency index, frame rate, frame delay, and frame loss are used as objective indicators, and user subjective ratings are used as subjective indicators; For clarity index, contrast, sharpness, and color saturation are used as objective indicators, and user subjective rating is used as subjective indicator; A deep learning algorithm is used to train and obtain a mapping model by taking the image features and grating parameter configuration of the displayed content as input and the fluency and clarity indicators as output; For dynamic video content, by analyzing its frame rate, motion speed, scene complexity, and matching the appropriate grating width and reflection angle parameter range, the smoothness and continuity of video playback are ensured; Genetic algorithm is used to optimize the parameter combination, and the average frame rate is taken as the primary optimization target of the genetic algorithm; For static chart content, by analyzing its edges, textures and colors, the appropriate grating width and reflection angle parameter range are matched to ensure the clarity and readability of the chart display; Genetic algorithm is used to optimize the parameter combination, and contrast is taken as the primary optimization target; For the text content, heuristic rules or search algorithms based on constraint satisfaction are used to allocate it to the remaining display area, and the grating width and reflection angle parameters are fine-tuned according to its fluency and clarity requirements.

10. The electronic ink screen 3D stereoscopic display method according to claim 2, characterized in that: The steps for adjusting the display content or parameters are specifically as follows: According to the physical characteristics and display requirements of the electronic ink screen, an adaptive partitioning algorithm is used to divide the screen into multiple rectangular display areas; Based on the semantic segmentation information of the displayed content, a deep learning model is used to perform pixel-level semantic annotation on the displayed image, identify different content types and display objects, and generate a semantic segmentation map, wherein the content types include text, images, and videos, and the display objects include titles, text, and icons; Combining the semantic segmentation map and pixel density information, the regions with similar semantics and similar pixel density are merged into one display area; The divided display areas are encoded and annotated to generate a region division map, and the position, size, pixel density and semantic category of each display area are extracted. Based on the extracted information, the parameters of each display area are optimized and dynamically adjusted.

Images