Electronic paper display enhancement method based on visual perception and dynamic prior information

Abstract

The invention relates to an electronic paper display enhancement method based on visual perception and dynamic prior information, and belongs to the technical field of display. The method comprises the following steps: firstly, quantizing electronic paper refreshing delay characteristics through a three-order gradient operator to generate display distortion priori, and dynamically predicting contrast enhancement parameters by using a convolutional network; secondly, a dual-channel unsupervised enhancement network is constructed, UNet-based ghosting suppression and anti-gamma transformation detail enhancement are performed respectively, and output is performed through adaptive weight fusion; and finally, designing a dynamic error diffusion system driven by visual perception, combining a frequency domain weighted filter and a spatial domain dynamic quantization threshold matrix, optimizing a refreshing sequence by adopting a snakelike scanning path, coding a processing result into a 16-order driving waveform, and controlling adaptive electronic paper hardware display through dynamic voltage. According to the method, the problems of gray scale distortion, edge blurring and ghosting effect of the high-dynamic-range image in electronic paper display are effectively solved, and the detail reduction degree and the visual comfort degree are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of display, and particularly relates to an electronic paper display enhancement method based on visual perception and dynamic prior information. BACKGROUND

[0002] With the rapid development of electronic paper display technology, its application in the fields of electronic readers, smart labels and the like is increasingly widespread. However, due to the electrophoretic display principle, the existing electronic paper device still has significant defects when presenting high dynamic range images: on the one hand, due to the limited gray level (usually only 16 levels of gray), the traditional error diffusion algorithm based on fixed threshold is easy to cause image distortion and edge blur, especially the gray level break phenomenon when expressing delicate texture; on the other hand, the unique ink particle response delay characteristic of electronic paper will introduce regional display distortion, and the existing method lacks the quantitative modeling ability of dynamic refresh ghosting. The current solution relies on supervised deep learning models, but the electronic paper display scene lacks ground truth (GT) training data, which seriously restricts the practicability of the algorithm.

[0003] It is worth noting that in the field of medical images, the research on X-ray image enhancement provides a breakthrough idea: first, an unsupervised distortion quantification is realized by constructing a noise estimation model (such as a high-order gradient prior); second, a reverse gamma transformation model is used to dynamically enhance the details of low-contrast regions, avoiding the limitations of manual parameter tuning; third, an innovative pseudo-GT generation mechanism solves the problem of missing constraints in unsupervised training. Although these technologies have achieved remarkable results in the field of X-rays, they have not been applied to electronic paper display optimization, especially the prior information guidance mechanism and the physical distortion characteristics of electronic paper have a high adaptation potential. SUMMARY

[0004] The purpose of the present application is to provide an electronic paper display enhancement method based on visual perception and dynamic prior information, which uses an unsupervised enhancement mechanism, combines the physical characteristics of electronic paper, and constructs a dual-channel prior information guidance model to significantly improve the gray level accuracy and edge detail retention capability of high dynamic range images, solving the problems of gray level distortion, edge blur and ghosting effect in traditional electronic paper display.

[0005] To achieve the above purpose, the technical scheme of the present application is as follows: an electronic paper display enhancement method based on visual perception and dynamic prior information, comprising the following steps:

[0006] Step S1, dynamically generating display distortion prior and contrast enhancement prior: extracting the gradient features of the input image, generating the display distortion prior in combination with the refresh delay characteristics of the electronic paper; at the same time, predicting the regional contrast enhancement prior;

[0007] Step S2, dual-channel unsupervised enhancement network processing: based on the display distortion prior generated in step S1, the contrast enhancement prior, distortion compensation and contrast enhancement are respectively performed;

[0008] Step S3, visual perception driven dynamic error diffusion and encoding generation: fusion of enhancement results to construct a visual weighted error diffusion system, and generate the required 16-order gray scale encoding for electronic paper display.

[0009] Further, step S1 includes:

[0010] Step S11, extract the gradient distribution and edge information of the image to be displayed, calculate the local distortion intensity by a third-order gradient operator, and calculate the display distortion prior σ EPD :

[0011]

[0012] where T refresh (x,y) is the historical refresh delay time at coordinates (x,y), T max = 300ms is the maximum refresh cycle threshold, represents the statistical average value of the third-order gradient of the input image I, represents the coefficient of the kth term in the third-order gradient calculation.

[0013] Step S12, construct a convolution network to dynamically generate a contrast enhancement prior γ:

[0014]

[0015] I is the input image matrix, K EPD is the electrophoretic response convolution kernel, and the convolution network structure ConvNet is a three-layer convolution layer: the first layer is 32 channels 3x3 convolution + ReLU, the second layer is 64 channels 3x3 convolution + ReLU, and the third layer is 1 channel 1x1 convolution + Sigmoid, with an output γ value range of [0.8, 2.2].

[0016] Step S13, calculate the local gray scale variation ΔG edge and generate the joint optimization weight ω:

[0017]

[0018] where N = 5x5 is the size of the local neighborhood window, i and j represent the average absolute deviation of the pixel values in the local region centered at pixel point (x,y), and μ is the average value of the pixels in the window.

[0019] Further, step S2 includes:

[0020] Step S21, distortion compensation branch processing: σEPD Generate a 32-channel feature map through 3*3 convolution, and input the UNet network after splicing with the input image channel;

[0021] Step S22, contrast enhancement branch processing: using inverse gamma transformation model to process the image after distortion compensation, and adding voltage constraint;

[0022] Step S23, fusing the double-channel outputs of step S21 and step S22.

[0023] Further, step S3 comprises:

[0024] Step S31, constructing a visual weighted error diffusion system and generating a dynamic quantization threshold matrix;

[0025] Step S32, optimizing the refresh sequence by adopting a snake-shaped scanning path;

[0026] Step S33, converting the optimized image data into electronic paper display coding.

[0027] Further, the method is suitable for electronic paper equipment using electrophoretic display technology.

[0028] The application also provides an electronic paper display enhancement system based on visual perception and dynamic prior information, comprising a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor, when the processor executes the computer program instructions, the steps of any of the above-mentioned methods can be realized.

[0029] The application also provides a computer readable storage medium, which stores computer program instructions capable of being executed by a processor, when the processor executes the computer program instructions, the steps of any of the above-mentioned methods can be realized.

[0030] Compared with the prior art, the application has the following beneficial effects: the application effectively solves the core technical bottleneck in the field of electronic paper display caused by the lack of real value training data by innovatively using an electronic paper image unsupervised enhancement mechanism; the noise estimation model is converted into an electronic paper display distortion quantization model, which significantly suppresses the regional residual image effect; the dynamically generated contrast enhancement prior parameter γ adaptively matches the nonlinear response characteristics of the electronic paper, so that the gray scale transition smoothness is substantially improved; the double-channel network structure inherits the advantages of the traditional error diffusion algorithm while significantly improving the visual fidelity of the edge details; the self-constraint mechanism based on pseudo-real values makes the unsupervised training effect approach the performance of the supervised model, and the detail restoration degree and visual comfort of the high dynamic range image in the electronic paper display are comprehensively optimized. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1The whole flow chart of the method of the application. DETAILED DESCRIPTION

[0032] The technical solutions of the application will be described in detail below with reference to the drawings.

[0033] The application provides an electronic paper display enhancement method based on visual perception and dynamic prior information, which comprises the following steps: EPD And a convolution network is used to dynamically predict a contrast enhancement parameter gamma; secondly, a double-channel unsupervised enhancement network is constructed to respectively perform residual image suppression based on UNet and detail enhancement through inverse gamma transformation, and an adaptive weight omega is used to fuse and output; finally, a visual perception driven dynamic error diffusion system is designed, a frequency domain weighting filter H(u, v) and a spatial domain dynamic quantization threshold matrix are combined, a snake-shaped scanning path is adopted to optimize a refresh sequence, a processing result is encoded into a 16-order driving waveform, and dynamic voltage control is adopted to adapt to electronic paper hardware display.

[0034] The following is a specific implementation process of the application.

[0035] The application provides an electronic paper display enhancement method based on visual perception and dynamic prior information, and the whole working process is as shown in the figure Figure 1 , and comprises the following steps:

[0036] Step S1, dynamically generating display distortion prior and contrast enhancement prior: gradient features of an input image are extracted, display distortion strength is generated in combination with electronic paper refresh delay characteristics, and regional contrast enhancement parameters are predicted;

[0037] Step S2, double-channel unsupervised enhancement network processing: based on the prior information generated in step S1, distortion compensation and contrast enhancement are respectively performed;

[0038] Step S3, visual perception driven dynamic error diffusion and encoding generation: a visual weighted error diffusion system is constructed by fusing and enhancing the results, and 16-order gray scale coding required for electronic paper display is generated.

[0039] In the embodiment, the step S1 is to generate dynamic prior information for a to-be-displayed image: image gradient features are extracted, display distortion strength parameters and contrast enhancement parameters are generated in combination with electronic paper physical characteristics, and the step specifically comprises the following steps:

[0040] Step S11, gradient distribution and edge information of the to-be-displayed image are extracted, local distortion strength is calculated through a three-order gradient operator, and display distortion prior sigma EPD is calculated in combination with pixel refresh delay time.

[0041]

[0042] where T refresh (x, y) is the historical refresh delay time at coordinates (x, y), T max = 300 ms is the maximum refresh period threshold, represents the third-order gradient statistical average of the input image I, represents the coefficient of the kth term in the third-order gradient calculation.

[0043] Step S12, construct a convolutional network to dynamically generate a contrast enhancement prior γ:

[0044]

[0045] I is an input image matrix, K EPD is an electrophoretic response convolution kernel, and the convolutional network structure ConvNet is a three-layer convolutional layer: a first layer of 32 channels 3×3 convolution + ReLU, a second layer of 64 channels 3×3 convolution + ReLU, and a third layer of 1 channel 1×1 convolution + Sigmoid, outputting a γ value range [0.8, 2.2].

[0046] Step S13, calculate the local gray level change ΔG edge and generate a joint optimization weight ω:

[0047]

[0048] where N = 5×5 is the size of the local neighborhood window, i and j represent the average absolute deviation of the pixel values in the local region centered at (x, y), and μ is the mean of the pixels in the window.

[0049] Next, the step S2, based on the generated dual-channel prior information, respectively performs distortion compensation and contrast enhancement processing, specifically including the following steps:

[0050] Step S21, distortion compensation branch processing: σ EPD is generated by 3×3 convolution to generate a 32-channel feature map, which is concatenated with the input image channel and input into the UNet network:

[0051] I denoise = UNet(Concat(I,Conv 3×3 (σ EPD ))) The UNet network structure contains 4 layers of down-sampling and 4 layers of up-sampling, and each layer has a convolution kernel size of 3×3 and a channel number of 64-128-256-512; I denoise is the output image after distortion compensation, Conv 3×3 is a 3×3 convolution layer that expands the scalar σ EPD into a 32-channel feature map, and Concat represents a channel concatenation operation that concatenates the original image I with the feature map.

[0052] Step S22, contrast enhancement branch processing: process the distortion compensation image with inverse gamma transformation model, and add electronic paper driving voltage constraint:

[0053] O enhance = min(max(1+(1-I denoise ) γ ,V min ),V max )

[0054] wherein V min = 0.1V, V max = 15V is the voltage range allowed by the electronic paper driving chip;

[0055] Step S23, dual-channel output fusion: weight and fuse the distortion compensation result and the contrast enhancement result:

[0056] I fused = ω·I denoise +(1-ω)·O enhance

[0057] Further, the step S3, dynamic error diffusion driven by visual perception: input the fused image into the error diffusion system, and combine the human visual characteristics to optimize the display effect, specifically including the following steps:

[0058] Step S31, constructing a visual weighting filter H(u,v) to quantify the visual error of the HVS sensitive area:

[0059]

[0060] wherein (u,v) represents the frequency domain spatial frequency coordinates, W EPD (x,y) is the electronic paper area weight: 1.2 for the gray transition area, and 0.8 for the flat area;

[0061] Generate a dynamic quantization threshold matrix, and set the threshold value adaptively according to the image area characteristics:

[0062]

[0063] and use an improved Floyd-Steinberg error diffusion algorithm:

[0064]

[0065] wherein f(x,y) is the original input pixel value, f * (x,y) is the pixel value after error diffusion optimization, β = 0.7 is the edge enhancement weight, and G *(x, y) is the gradient amplitude calculated by Sobel operator, W(i, j) is the Floyd-Steinberg error diffusion weight matrix, and e(x-i, y-j) is the accumulated quantization error of adjacent pixels.

[0066] In step S32, the pixel processing order is optimized by using a "snake" scanning path, in order to avoid particle flip inertia, odd rows are scanned from left to right with a step size Δt=15 ms, and even rows are scanned from right to left.

[0067] In step S33, the optimized image data is converted into an electronic paper display code, and a quantization process is performed according to the 16-level grayscale mapping specification of the electronic paper, to generate a code format that can be directly used to drive display.

[0068] The application also provides an electronic paper display enhancement system based on visual perception and dynamic prior information, which comprises a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor, and when the processor executes the computer program instructions, the steps of any of the above methods can be implemented.

[0069] The application also provides a computer readable storage medium, which stores computer program instructions capable of being executed by a processor, and when the processor executes the computer program instructions, the steps of any of the above methods can be implemented.

[0070] The above is the preferred embodiment of the application, and any changes made according to the technical solutions of the application, as long as the generated function does not exceed the scope of the technical solutions of the application, are all within the protection scope of the application.

Claims

1. A method for enhancing electronic paper displays based on visual perception and dynamic prior information, characterized in that, Includes the following steps: Step S1: Dynamically generate display distortion prior and contrast enhancement prior: Extract gradient features from the input image and combine them with the refresh delay characteristics of electronic paper to generate display distortion prior; at the same time, predict the region contrast enhancement prior. Step S2, Dual-channel unsupervised enhancement network processing: Based on the display distortion prior and contrast enhancement prior generated in step S1, distortion compensation and contrast enhancement are performed respectively; Step S3, Visual Perception-Driven Dynamic Error Diffusion and Encoding Generation: The fusion enhancement results are used to construct a visual weighted error diffusion system and generate the 16-level grayscale encoding required for electronic paper display.

2. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 1, characterized in that, Step S1 includes: Step S11: Extract the gradient distribution and edge information of the image to be displayed, calculate the local distortion intensity using the third-order gradient operator, and calculate the prior display distortion σ by combining the pixel refresh delay time. EPD ; Step S12: Construct a convolutional network to dynamically generate contrast-enhancing prior γ; Step S13: Calculate the local grayscale change ΔG edge And generate joint optimization weights ω.

3. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 2, characterized in that, Displaying the distorted prior σ EPD It is expressed as follows: Where T refresh (x,y) represents the historical refresh delay time at coordinates (x,y), T max =300ms is the maximum refresh cycle threshold. This represents the statistical average of the third gradient of the input image I. This represents the coefficient of the k-th term in the third-order gradient calculation.

4. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 3, characterized in that, The prior γ for contrast enhancement is represented as follows: I is the input image matrix, K EPD The convolutional network structure ConvNet consists of three convolutional layers: the first layer is a 32-channel 3×3 convolution + ReLU, the second layer is a 64-channel 3×3 convolution + ReLU, and the third layer is a 1-channel 1×1 convolution + Sigmoid. The output γ value ranges from [0.8, 2.2].

5. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 4, characterized in that, Local grayscale change ΔG edge The joint optimization weight ω is expressed as follows: Where N = 5 × 5 is the size of the local neighborhood window, i and j represent the average absolute deviation of pixel values ​​in the local region centered at pixel (x, y), and μ is the mean pixel value within the window.

6. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 2, characterized in that, Step S2 includes: Step S21, Distortion Compensation Branch Processing: σ EPD A 32-channel feature map is generated by 3×3 convolution, which is then concatenated with the input image channels and input into the UNet network. Step S22, Contrast Enhancement Branch Processing: The image after distortion compensation is processed using an inverse gamma transform model, and voltage constraints are added; Step S23: Combine the dual-channel outputs from steps S21 and S22.

7. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 6, characterized in that, Step S3 includes: Step S31: Construct a visual weighted error diffusion system and generate a dynamic quantization threshold matrix; Step S32: Optimize the refresh sequence using a serpentine scanning path; Step S33: Convert the optimized image data into electronic paper display encoding.

8. The electronic paper display enhancement method based on visual perception and dynamic prior information according to claim 1, characterized in that, Suitable for electronic paper devices that employ electrophoretic display technology.

9. An electronic paper display enhancement system based on visual perception and dynamic prior information, characterized in that, It includes a memory, a processor, and computer program instructions stored in the memory and executable by the processor, which, when executed by the processor, enable the implementation of the steps of the method as described in any one of claims 1-8.

10. A computer-readable storage medium having stored thereon computer program instructions executable by a processor, wherein when the processor executes the computer program instructions, it is able to implement the steps of the method as described in any one of claims 1-8.

Images