LCD screen resolution improving method based on virtual pixel mapping

Abstract

The invention discloses an LCD screen resolution improving method based on virtual pixel mapping. The LCD screen resolution improving method comprises the following steps that low-resolution original image data needing to be displayed on an LCD screen is collected and preprocessed; obtaining physical parameter information of the LCD screen, and generating a virtual pixel mapping rule; constructing and training an improved sub-pixel convolutional neural network to obtain an improved sub-pixel convolutional neural network model; inputting the standardized low-resolution input image into the improved sub-pixel convolutional neural network model, and outputting an LCD screen high-resolution feature image; performing multi-branch virtual pixel channel recombination on the LCD screen high-resolution feature image to obtain a virtual pixel recombination image; an adaptive sub-pixel fusion mechanism is adopted to obtain an optimized virtual pixel image; and based on the optimized virtual pixel image, driving an LCD screen to display the image content of which the resolution is improved. According to the method, a virtual pixel mapping and deep learning fusion method is adopted, and the LCD screen resolution is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing and display, and particularly relates to an LCD screen resolution improvement method based on virtual pixel mapping. BACKGROUND

[0002] With the wide application of liquid crystal display screens (LCD) in the fields of advertising, information terminals and mobile devices, improving the display resolution of LCD screens has become the focus of the industry. Existing resolution improvement methods mainly include improving the physical pixel density of the panel and using image processing algorithms for super-resolution reconstruction. The former can effectively improve the picture accuracy, but often accompanied by rising production costs and increasing manufacturing difficulty. The latter, such as bilinear interpolation and convolutional neural network algorithm, can improve the display effect of low-resolution images at low cost and is widely used in actual products.

[0003] However, the current super-resolution method based on deep learning usually only processes pixel-level images, ignoring the structural characteristics of the physical sub-pixel arrangement of the LCD screen, and fails to achieve accurate mapping of virtual pixels and actual physical sub-pixels. This results in insufficient detail restoration and insufficient color consistency of the image after resolution improvement when displayed on the actual LCD. The existing technology also lacks a mechanism for adaptive optimization and weight adjustment for different LCD panel physical parameters and sub-pixel structures, making it difficult to meet the actual needs of high resolution and high display quality.

[0004] Therefore, how to provide an LCD screen resolution improvement method based on virtual pixel mapping is a problem that those skilled in the art need to solve. SUMMARY

[0005] One object of the present application is to provide an LCD screen resolution improvement method based on virtual pixel mapping. The present application combines the physical sub-pixel structure of the LCD screen with the improved sub-pixel convolutional neural network of deep learning, and realizes high-precision resolution improvement of low-resolution images on the LCD screen through virtual pixel mapping, multi-branch virtual pixel channel reorganization and adaptive sub-pixel fusion mechanism. This method can customize mapping and weight distribution according to the sub-pixel arrangement characteristics of different LCD screens, and has the advantages of high detail restoration, good color consistency and strong display adaptability.

[0006] According to an LCD screen resolution improvement method based on virtual pixel mapping according to an embodiment of the present application, the method comprises the following steps: Collect and pre-process low-resolution original image data that needs to be displayed on the LCD screen to obtain a standardized low-resolution input image; Obtain the physical parameter information of the LCD screen to generate a virtual pixel mapping rule; According to the sub-pixel arrangement mode of the LCD screen and the virtual pixel mapping rule, an improved sub-pixel convolutional neural network model is constructed and trained; The normalized low-resolution input image is input into the improved sub-pixel convolutional neural network model, and an LCD screen high-resolution feature image is output; According to the virtual pixel mapping rule, the LCD screen high-resolution feature image is subjected to multi-branch virtual pixel channel reorganization, and a virtual pixel reorganized image is obtained; An adaptive sub-pixel fusion mechanism is adopted, and the sub-pixel weight distribution is dynamically adjusted according to the content features of the virtual pixel reorganized image and the sub-pixel arrangement mode of the LCD screen, and an optimized virtual pixel image is obtained; The optimized virtual pixel image is output to the display driving system of the LCD screen according to the virtual pixel mapping rule, and the LCD screen displays the image content with improved resolution.

[0007] Optionally, the normalized low-resolution input image is a multi-channel image data matrix consistent with the size of the LCD screen display area, which is normalized and size-standardized.

[0008] Optionally, the obtaining of the physical parameter information of the LCD screen and the generation of the virtual pixel mapping rule specifically include: The resolution parameters of the LCD screen are obtained, and the resolution parameters of the LCD screen include the height pixel number and the width pixel number of the LCD screen display area; The physical size parameters of the LCD screen are obtained, and the actual physical size and position distribution of each physical pixel and each sub-pixel in the display area are calculated according to the corresponding relationship between the physical size and the resolution of the LCD screen; The sub-pixel arrangement mode of the LCD screen is obtained, and the arrangement order, type and spatial position of each sub-pixel in each physical pixel are determined; According to the resolution parameters of the LCD screen, the physical size parameters of the LCD screen and the sub-pixel arrangement mode of the LCD screen, the mapping relationship between each virtual pixel and the physical sub-pixel of the LCD screen is established, and the virtual pixel mapping rule is formed.

[0009] Optionally, the construction and training of the improved sub-pixel convolutional neural network model according to the sub-pixel arrangement mode of the LCD screen and the virtual pixel mapping rule specifically include: The improved sub-pixel convolutional neural network model is constructed, and the improved sub-pixel convolutional neural network model includes a sub-pixel perception convolution module, a deep feature extraction module, a channel expansion convolution module and a sub-pixel convolution rearrangement module; According to the virtual pixel mapping rule, the initial parameters of the convolution kernel of the sub-pixel perception convolution module are determined; The deep feature extraction module is composed of a plurality of convolutional layers and an activation function stack, and is configured to extract high-dimensional feature representation of normalized image data as input, and output a deep feature map; The channel expansion convolution module is configured to expand the number of channels of the deep feature map to obtain a channel expansion feature map. The sub-pixel convolution rearrangement module is configured to rearrange the channel expansion feature map into a high-resolution virtual pixel image according to space. The improved sub-pixel convolutional neural network model is trained by defining a sub-pixel perception loss function for the sub-pixel arrangement of the LCD screen, with a normalized low-resolution input image as input and a corresponding high-resolution target image as expected output. The improved sub-pixel convolutional neural network model is iteratively trained using the gradient descent method until the value of the sub-pixel perception loss function reaches the set convergence condition, and the trained improved sub-pixel convolutional neural network model is output.

[0010] Optionally, the initial parameters of the sub-pixel perception convolution module are determined according to the virtual pixel mapping rule, and specifically include: According to the virtual pixel mapping rule, the spatial distribution characteristics of all types of sub-pixels in the display area are counted and labeled in combination with the resolution, physical size and sub-pixel arrangement of the LCD screen, and a sub-pixel spatial index map is obtained. According to the sub-pixel spatial index map, a plurality of convolution kernels are designed for each type of sub-pixel arrangement, and the convolution kernel size parameters and the receptive field parameters are obtained. According to the sub-pixel spatial index map, the convolution kernel size parameters and the receptive field parameters, the number of each group of convolution kernels is set to be consistent with the number of LCD screen sub-pixel structure types, and the convolution kernel number parameters are obtained. According to the virtual pixel mapping rule, the physical space position of the input virtual pixel in the LCD display area is one-to-one mapped with the corresponding physical sub-pixel unit, and the virtual pixel spatial mapping relationship is obtained. According to the virtual pixel spatial mapping relationship, the weight distribution of each convolution kernel is initialized, and the initial weight distribution parameters of the convolution kernel are obtained. For the LCD screen with non-linear sub-pixel arrangement, the receptive field and the initial weight distribution of the convolution kernel are adjusted according to the virtual pixel spatial mapping relationship, and the non-linear arrangement convolution kernel parameters are obtained. The convolution kernel number parameters, the convolution kernel size parameters, the receptive field parameters, the initial weight distribution parameters of the convolution kernel and the non-linear arrangement convolution kernel parameters are output.

[0011] Optionally, the channel expansion convolution module expands the number of channels of the deep feature map to obtain a channel expansion feature map, which specifically includes: The deep feature map is received, original channel number, spatial size and data format of the deep feature map are counted, and deep feature map parameter information is obtained; According to the LCD screen resolution improvement target, the virtual pixel mapping rule and the spatial reorganization requirement of the sub-pixel convolution rearrangement module, the required target output channel number is calculated, and a target output channel number parameter is obtained; Based on the deep feature map parameter information and the target output channel number parameter, the number and size of the convolution kernel of the channel expansion convolution module are set, and a channel expansion convolution kernel configuration parameter is obtained; The deep feature map is input into the channel expansion convolution module, and the preset convolution kernel configuration is used to expand and combine all channels of the deep feature map, and the channel expansion feature map data is output.

[0012] Optionally, the input of the standardized low-resolution input image into the improved sub-pixel convolutional neural network model and the output of the LCD screen high-resolution feature image specifically include: The standardized low-resolution input image is input into the improved sub-pixel convolutional neural network model trained according to the LCD screen sub-pixel arrangement mode and the virtual pixel mapping rule; In the sub-pixel perception convolution module, the input image is preliminarily feature extracted according to the sub-pixel arrangement characteristics of the LCD screen, and a sub-pixel feature representation is obtained; In the deep feature extraction module, the sub-pixel feature representation is further subjected to multi-layer convolution processing and high-dimensional feature extraction, and a deep feature map is obtained; In the channel expansion convolution module, the deep feature map is expanded in the channel dimension according to the resolution improvement requirement, and a channel expansion feature map is obtained; In the sub-pixel convolution rearrangement module, the channel expansion feature map is spatially reorganized, and the channel information is redistributed to the spatial dimension, and the LCD screen high-resolution feature image is output.

[0013] Optionally, the multi-branch virtual pixel channel reorganization of the LCD screen high-resolution feature image according to the virtual pixel mapping rule to obtain a virtual pixel reorganization image specifically includes: The LCD screen high-resolution feature image is received, and high-resolution feature channel data is obtained; According to the virtual pixel mapping rule, the correspondence relationship between each channel of the high-resolution feature channel data and the LCD screen physical sub-pixel type, arrangement order and spatial position is analyzed, and a channel-sub-pixel mapping relationship is obtained; The high-resolution feature channel data is divided into a plurality of virtual pixel reorganization branches according to the channel-sub-pixel mapping relationship, and virtual pixel reorganization branch data is obtained; For each virtual pixel reorganization branch, according to the virtual pixel mapping rule, each color component of the input virtual pixel is distributed to the corresponding physical sub-pixel unit according to the spatial mapping and arrangement requirements, and a branch weighted reorganization result is obtained; For different sub-pixel arrangement modes of the LCD screen, the spatial combination order and weight ratio of each virtual pixel reorganization branch are dynamically adjusted, all branch weighted reorganization results are integrated, and a virtual pixel reorganization image is formed.

[0014] Optionally, the adaptive sub-pixel fusion mechanism is used to dynamically adjust the sub-pixel weight distribution according to the content features of the virtual pixel reorganization image and the sub-pixel arrangement mode of the LCD screen, and an optimized virtual pixel image is obtained, which specifically includes: The virtual pixel reorganization image is received, and the spatial distribution, color composition, edge feature, and texture distribution of the virtual pixel reorganization image are analyzed to obtain image content feature data; According to the sub-pixel arrangement mode of the LCD screen and in combination with the virtual pixel mapping rule, an initial weight distribution scheme is set for each virtual pixel and the corresponding physical sub-pixel region, and initial sub-pixel weight parameters are obtained; According to the content feature data of the virtual pixel reorganization image, an adaptive weighting mechanism is used to dynamically adjust the sub-pixel weight parameters in the spatial region, and adaptive adjustment weight parameters are obtained; Through a content-aware fusion algorithm, the initial sub-pixel weight parameters and the adaptive adjustment weight parameters are jointly weighted, the optimal weight distribution of each color channel component of each virtual pixel on different physical sub-pixel units is calculated, and a final sub-pixel weight distribution matrix is obtained; According to the final sub-pixel weight distribution matrix, the virtual pixel reorganization image is weighted and fused in each spatial position and color channel component, and an optimized virtual pixel image is output.

[0015] Optionally, the optimized virtual pixel image is output to the display driving system of the LCD screen according to the virtual pixel mapping rule, and the display driving system of the LCD screen displays the image content with improved resolution, which specifically includes: The optimized virtual pixel image is received, and according to the virtual pixel mapping rule, each virtual pixel and each color channel component in the optimized virtual pixel image is matched to a specific sub-pixel unit in the physical display region of the LCD screen, forming a driving data mapping relationship; According to the driving data mapping relationship, a driving data packet is generated, and the driving data packet includes the final display brightness or voltage control value of each physical sub-pixel; The driving data packet is output to the display driving system of the LCD screen, and the physical sub-pixels of the LCD screen are driven to display according to the optimized virtual pixel image.

[0016] The beneficial effects of the present application are: The present application can realize high-precision mapping and recombination of a low-resolution input image according to the physical parameters and the subpixel arrangement mode of an LCD screen by introducing a virtual pixel mapping rule in a resolution enhancement process, thereby ensuring one-to-one correspondence between virtual pixels and actual physical subpixels. This mapping mechanism effectively avoids the structural mismatch problem existing in the actual display of the LCD in the traditional pixel-level super-resolution algorithm, and improves the adaptability of the resolution enhancement result on the physical screen.

[0017] By designing an improved subpixel convolutional neural network structure, the present application can fully extract and utilize the spatial detail information of the image, combine virtual pixel recombination with the end-to-end feature optimization process of deep learning, and realize adaptive feature learning for different LCD subpixel arrangement modes. After the high-resolution feature image output by the network is recombined through a multi-branch channel and an adaptive subpixel fusion mechanism, the subpixel-level detail restoration and color correction can be realized, thereby effectively improving the definition and consistency of the final display effect.

[0018] The method of the present application does not need to change the hardware structure of the LCD screen, and has strong universality and actual deployability. By flexibly configuring for different panel structures and display requirements, the requirements for high resolution and high display quality in application scenarios such as advertising display, information terminal, industrial monitoring, etc. can be met, which helps to improve the overall visual performance of the LCD display terminal. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application and explain the principles of the present application, and do not constitute a limitation of the present application. In the drawings:

[0020] Fig. 1 A flowchart of an LCD screen resolution enhancement method based on virtual pixel mapping according to the present application; Fig. 2 A schematic diagram of an improved subpixel convolutional neural network structure in an LCD screen resolution enhancement method based on virtual pixel mapping according to the present application; Fig. 3 A processing schematic diagram of an adaptive subpixel fusion mechanism in an LCD screen resolution enhancement method based on virtual pixel mapping according to the present application. DETAILED DESCRIPTION

[0021] The present application will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, and only illustrate the basic structure of the present application in a schematic manner, and therefore only show the components related to the present application.

[0022] REFERENCE Figs. 1-3The application discloses a LCD screen resolution improvement method based on virtual pixel mapping, and comprises the following steps: Collect and pre-process low-resolution original image data to be displayed on the LCD screen to obtain standardized low-resolution input images; Obtain physical parameter information of the LCD screen to generate a virtual pixel mapping rule; According to the sub-pixel arrangement mode of the LCD screen and the virtual pixel mapping rule, an improved sub-pixel convolutional neural network is constructed and trained to obtain an improved sub-pixel convolutional neural network model; The standardized low-resolution input images are input into the improved sub-pixel convolutional neural network model to output high-resolution feature images of the LCD screen; According to the virtual pixel mapping rule, the high-resolution feature images of the LCD screen are subjected to multi-branch virtual pixel channel recombination to obtain virtual pixel recombination images; An adaptive sub-pixel fusion mechanism is adopted to dynamically adjust sub-pixel weight distribution according to the content features of the virtual pixel recombination images and the sub-pixel arrangement mode of the LCD screen to obtain optimized virtual pixel images; The optimized virtual pixel images are output to the display driving system of the LCD screen according to the virtual pixel mapping rule to drive the LCD screen to display the image content with improved resolution.

[0023] In the embodiment, the collecting and pre-processing of the low-resolution original image data to be displayed on the LCD screen to obtain the standardized low-resolution input images specifically comprises the following steps: Original image data to be displayed on the LCD screen is obtained through an image acquisition device or an input interface, wherein the original image data is low-resolution and multi-channel image data, and comprises RGB channel information consistent with the color displayed by the LCD screen; The original image data is subjected to normalization processing, and the channel values of each pixel point are uniformly converted into a standardized numerical range of 0 to 1 to adapt to the color display characteristics of the LCD screen, thereby obtaining normalized image data; The normalized image data is subjected to size adjustment processing, and the spatial size of the image is adjusted to be completely consistent with the size of the display area of the LCD screen, and pixel resampling is performed by using an interpolation algorithm to obtain image data corresponding to the physical display area size of the LCD screen; The standardized low-resolution input images are output, wherein the standardized low-resolution input images are multi-channel image data matrices which are consistent with the size of the display area of the LCD screen and have been subjected to normalization and size standardization processing, the channel value range of each pixel point is 0 to 1, the original image content can be completely expressed in terms of spatial distribution, color component and intensity, and the input requirements of a subsequent deep learning model for LCD screen resolution improvement are met.

[0024] In the embodiment, the obtaining of the physical parameter information of the LCD screen specifically comprises: obtaining a resolution parameter of the LCD screen, the resolution parameter of the LCD screen comprising a height pixel number and a width pixel number of a display area of the LCD screen, and being used to determine a corresponding relationship between a virtual pixel and an actual physical pixel; obtaining a physical size parameter of the LCD screen, the physical size parameter of the LCD screen comprising a physical height and a physical width of the display area, and being used to calculate an actual physical size and a position distribution of each physical pixel and each sub-pixel in the display area according to a corresponding relationship between the physical size and the resolution; obtaining a sub-pixel arrangement mode of the LCD screen, the sub-pixel arrangement mode of the LCD screen comprising but not limited to RGB, BGR and PenTile, and being used to explicitly determine an arrangement order, a type and a spatial position of each sub-pixel in each physical pixel; establishing a mapping relationship between each virtual pixel and a physical sub-pixel of the LCD screen according to the resolution parameter of the LCD screen, the physical size parameter of the LCD screen and the sub-pixel arrangement mode of the LCD screen, comprising explicitly determining a physical position of each input virtual pixel in the display area of the LCD screen and corresponding physical pixel and sub-pixel units, defining a specific way in which each color component (such as red, green and blue) of each virtual pixel is allocated to an actual physical sub-pixel in combination with the sub-pixel arrangement mode, and if the sub-pixel arrangement is a nonlinear structure (such as PenTile), determining a weighted allocation ratio of each virtual pixel component to multiple physical sub-pixels according to an actual arrangement rule through a lookup table, a weighted average or an interpolation algorithm; forming a virtual pixel mapping rule, the virtual pixel mapping rule comprising a mapping matrix, a weight allocation table or a spatial arrangement algorithm between the input virtual pixel and the actual physical sub-pixel, and being used to allocate processed image data to corresponding sub-pixels of the LCD screen according to the mapping relationship, so as to realize high-precision resolution improvement and detail enhancement.

[0025] In the embodiment, the constructing and training of the improved sub-pixel convolutional neural network model according to the sub-pixel arrangement mode of the LCD screen and the virtual pixel mapping rule specifically comprises: constructing an improved sub-pixel convolutional neural network model, the improved sub-pixel convolutional neural network model comprising a sub-pixel perception convolution module, a deep feature extraction module, a channel expansion convolution module and a sub-pixel convolution rearrangement module; determining an initial parameter of a convolution kernel of the sub-pixel perception convolution module according to the virtual pixel mapping rule, the initial parameter of the convolution kernel comprising a number, a size and an initial weight distribution of the convolution kernel, so that an initial structure thereof is adapted to the sub-pixel arrangement mode of the LCD screen; The deep feature extraction module consists of multiple convolutional layers and stacked activation functions, used to extract high-dimensional feature representations from the input normalized image data and output deep feature maps. The channel-expanded convolution module expands the number of channels in the deep feature map to obtain a channel-expanded feature map, which is used as the input to the subsequent sub-pixel convolution rearrangement module. The subpixel convolutional rearrangement module uses pixel rearrangement operations to spatially reorganize the channel extended feature map into a high-resolution virtual pixel image, thereby improving resolution and enhancing details. Using a standardized low-resolution input image as input and the corresponding high-resolution target image as the desired output, an improved sub-pixel convolutional neural network model is trained by defining a sub-pixel perceptual loss function tailored to the sub-pixel arrangement of LCD screens. ; in, For sub-pixel perception loss function, The total number of virtual pixels in the training samples. For the index of the virtual pixel, For color channel components, It is a red channel. For the blue channel, For green channel, For the high-resolution target image, the first The virtual pixel of the first The true values ​​of the color channel components To improve the image output of subpixel convolutional neural network models, the first... The virtual pixel of the first Predicted values ​​for color components; This formula is essentially the Mean Squared Error (MSE) loss function, a classic loss function in machine learning, deep learning, and image processing. Its original form is as follows: ; in, For the sample size, For sample index, For the true value, This is a predicted value; Based on MSE, this patent combines the structural characteristics of virtual pixels and color channels in this method, refines the loss metric into the summation of all virtual pixels and all color channel components, calculates the squared error between the prediction and the target on each virtual pixel and each color channel component, and normalizes it according to the total number of virtual pixels. In the formula for the sub-pixel perception loss function, , All are the numerical values of the pixel color channel components of the image, dimensionless, and the operations in the formula do not involve dimension conversion, therefore, the physical dimensions of both sides of the formula are consistent and conform to common sense; The improved sub-pixel convolutional neural network model is iteratively trained using a gradient descent method until the value of the sub-pixel perception loss function reaches a set convergence condition, and the trained improved sub-pixel convolutional neural network model is output.

[0026] In this embodiment, the determination of the initial parameters of the convolution kernel of the sub-pixel perception convolution module according to the virtual pixel mapping rule specifically includes: According to the virtual pixel mapping rule, the spatial distribution characteristics of all types of sub-pixels in the display area are counted and labeled in combination with the resolution, physical size and sub-pixel arrangement of the LCD screen, and a sub-pixel spatial index map is obtained; According to the sub-pixel spatial index map, a plurality of groups of convolution kernels are designed for each type of sub-pixel arrangement, the size and receptive field of each group of convolution kernels are determined, and the convolution kernel size parameters and receptive field parameters are obtained; According to the sub-pixel spatial index map, the convolution kernel size parameters and the receptive field parameters, the number of each group of convolution kernels is set to be consistent with the number of LCD screen sub-pixel structure types, and the convolution kernel number parameters are obtained; According to the virtual pixel mapping rule, the physical space position of the input virtual pixel in the LCD display area is one-to-one mapped with the corresponding physical sub-pixel unit, and a virtual pixel spatial mapping relationship is obtained; According to the virtual pixel spatial mapping relationship, the weight distribution of each convolution kernel is initialized, and the initial weight distribution parameters of the convolution kernel are obtained, which are used to make the convolution kernel focus on the corresponding sub-pixel type and the surrounding sub-pixel characteristics; For the LCD screen with non-linear sub-pixel arrangement, according to the virtual pixel spatial mapping relationship, the receptive field and the initial weight distribution of the convolution kernel are adjusted by using a lookup table, interpolation or weighted average algorithm, and non-linear arrangement convolution kernel parameters are obtained; The convolution kernel number parameters, the convolution kernel size parameters, the receptive field parameters, the initial weight distribution parameters of the convolution kernel and the non-linear arrangement convolution kernel parameters are output, and the above parameters are loaded into the sub-pixel perception convolution module, so that the initial structure of the sub-pixel perception convolution module can realize accurate feature extraction for different LCD screen sub-pixel arrangement modes and spatial mapping rules.

[0027] In this embodiment, the channel expansion convolution module expands the channel number of the deep feature map to obtain a channel expansion feature map, which specifically includes: The deep feature map is received, the original channel number, spatial size and data format of the deep feature map are counted, and deep feature map parameter information is obtained; According to the LCD screen resolution improvement target, the virtual pixel mapping rule and the spatial reorganization requirement of the sub-pixel convolution rearrangement module, the required target output channel number is calculated to obtain the target output channel number parameter; The calculation of the required target output channel number includes determining the number of color channels contained in the deep feature map itself, for example, a color image usually has three channels; then, combined with the magnification of the resolution improvement, according to the requirement that each input channel needs to be split into multiple output channels after spatial expansion, the total number of channels that need to be finally output is calculated; The channel expansion convolution module will set the number and structure of the convolution kernel according to this calculation result, so that the output feature map can meet the requirements of subsequent pixel spatial rearrangement and virtual pixel mapping in the number of channels; in this way, after the channel expansion processing, the obtained feature map not only has a higher number of channels, but also strictly matches the input specifications of the subsequent sub-pixel convolution rearrangement module, thereby ensuring that the finally generated high-resolution virtual pixel image can be consistent with the actual display requirements of the LCD screen, realizing effective resolution improvement and spatial detail restoration; Based on the deep feature map parameter information and the target output channel number parameter, the number and size of the convolution kernel of the channel expansion convolution module are set, so that each group of convolution kernels can map and enhance the features related to the LCD virtual pixel structure in the channel dimension to obtain the channel expansion convolution kernel configuration parameter; The deep feature map is input into the channel expansion convolution module, and the preset convolution kernel configuration is used to expand and combine all channels of the deep feature map, and a channel expansion feature map with consistent output channel number and spatial reorganization requirement is output to obtain the channel expansion feature map data; The output channel expansion feature map data is used as the input basis for the subsequent sub-pixel convolution rearrangement module to reorganize the spatial dimension and generate a high-resolution virtual pixel image.

[0028] In this embodiment, the input of the standardized low-resolution input image into the improved sub-pixel convolutional neural network model and the output of the LCD screen high-resolution feature image specifically include: The standardized low-resolution input image is received, which is a multi-channel image data processed by normalization and size standardization, consistent with the display area size of the LCD screen; The standardized low-resolution input image is input into the improved sub-pixel convolutional neural network model trained according to the LCD screen sub-pixel arrangement mode and the virtual pixel mapping rule, and sequentially passes through the sub-pixel perception convolution module, the deep feature extraction module, the channel expansion convolution module and the sub-pixel convolution rearrangement module to gradually process the input image data; In the sub-pixel perception convolution module, the input image is preliminarily feature extracted according to the sub-pixel arrangement characteristics of the LCD screen to obtain a sub-pixel feature representation; In the deep feature extraction module, further multi-layer convolution processing and high-dimensional feature extraction are performed on the sub-pixel feature representation to obtain a deep feature map; In the channel expansion convolution module, according to the resolution enhancement requirement, the deep feature map is expanded in the channel dimension to obtain a channel expansion feature map; In the sub-pixel convolution rearrangement module, the channel expansion feature map is spatially reorganized, and the channel information is redistributed to the spatial dimension to output an LCD screen high-resolution feature image, which has the characteristics of virtual pixel level resolution enhancement and spatial detail enhancement.

[0029] In the embodiment, the multi-branch virtual pixel channel rearrangement of the LCD screen high-resolution feature image according to the virtual pixel mapping rule to obtain a virtual pixel rearrangement image specifically includes: Receiving an LCD screen high-resolution feature image, which is a multi-channel image data output by the sub-pixel convolution rearrangement module, contains all color component information processed by spatial enhancement and channel expansion, and obtains high-resolution feature channel data; According to the virtual pixel mapping rule, the correspondence between each channel of the high-resolution feature channel data and the LCD screen physical sub-pixel type, arrangement order and spatial position is analyzed, the mapping matrix of the virtual pixel and the actual physical sub-pixel is determined, and the channel-sub-pixel mapping relationship is obtained; The high-resolution feature channel data is divided into a plurality of virtual pixel rearrangement branches according to the channel-sub-pixel mapping relationship, each branch corresponds to a sub-pixel structure or a specific physical area of the LCD screen, and virtual pixel rearrangement branch data is obtained; For each virtual pixel rearrangement branch, according to the virtual pixel mapping rule, the color components of the input virtual pixel are allocated to the corresponding physical sub-pixel unit according to the spatial mapping and arrangement requirements by using weight weighting, interpolation and table lookup, the weighted allocation of the virtual pixel color components to multiple physical sub-pixels is realized, and branch weighted rearrangement results are obtained; For different sub-pixel arrangement modes of the LCD screen (including but not limited to RGB, BGR, PenTile, and diagonal arrangement), the spatial combination order and weight proportion of each virtual pixel rearrangement branch are dynamically adjusted, all branch weighted rearrangement results are integrated, and a virtual pixel rearrangement image conforming to the LCD physical sub-pixel arrangement is formed; Outputting the virtual pixel rearrangement image, the data structure of the virtual pixel rearrangement image is one-to-one corresponding to the LCD screen physical sub-pixel arrangement, which is used as the input basis for subsequent adaptive sub-pixel fusion mechanism and driving display.

[0030] In this embodiment, the adaptive sub-pixel fusion mechanism dynamically adjusts the sub-pixel weight allocation based on the content features of the virtual pixel reconstructed image and the sub-pixel arrangement of the LCD screen to obtain an optimized virtual pixel image. Specifically, this includes: Receive virtual pixel reconstructed images, analyze the spatial distribution, color composition, edge features, and texture distribution of the virtual pixel reconstructed images, and obtain image content feature data; Based on the sub-pixel arrangement of the LCD screen and combined with the virtual pixel mapping rules, an initial weight allocation scheme is set for each virtual pixel and its corresponding physical sub-pixel region to obtain the initial sub-pixel weight parameters. For the content feature data of virtual pixel reconstructed images, an adaptive weighting mechanism is adopted to dynamically adjust the sub-pixel weight parameters within the spatial region, thereby enhancing the sensitivity of sub-pixel distribution in edge and texture change regions of the image and obtaining adaptively adjusted weight parameters. By using a content-aware fusion algorithm, the initial sub-pixel weight parameters and the adaptively adjusted weight parameters are jointly weighted to calculate the final weight allocation of each color channel component of each virtual pixel on different physical sub-pixel units, thus obtaining the final sub-pixel weight allocation matrix: ; in, For the first OK Column virtual pixels in the first Final subpixel weights on color channel components The weighted fusion coefficient, For the first OK Column virtual pixels in the first Initial subpixel weight parameters on color channel components For the first OK Column virtual pixels in the first Adaptive adjustment of weight parameters on color channel components This is the row index of the virtual pixel in the image. The column index of the virtual pixel in the image; The formula employs a weighted linear combination method commonly used in signal processing and image processing. Its original form is a linear weighted sum of two weight parameters. In this patent, the formula, combined with the specific needs of LCD screen resolution improvement and virtual pixel mapping, weights and fuses the initial weight parameters of the virtual pixel at each spatial location and color channel component with the adaptively adjusted weight parameters to form the final sub-pixel weight allocation. Both sides of the formula represent the sub-pixel weights of the virtual pixel at different spatial locations and color channels, and are dimensionless proportions, ensuring dimensional consistency and conforming to engineering common sense. According to the final sub-pixel weight distribution matrix, the virtual pixel reorganization image is weighted and fused on each spatial position and color channel component, and an optimized virtual pixel image is output, which has improved spatial detail performance and display uniformity, and meets the actual display requirements of the LCD screen.

[0031] In the embodiment, the display driving system for outputting the optimized virtual pixel image to the LCD screen according to the virtual pixel mapping rule specifically includes: The optimized virtual pixel image is received, which is image data processed by the adaptive sub-pixel fusion mechanism, and the accurate weight distribution between the virtual pixel and the physical sub-pixel has been realized; According to the virtual pixel mapping rule, each virtual pixel and each color channel component in the optimized virtual pixel image is matched to a specific sub-pixel unit in the physical display area of the LCD screen, forming a driving data mapping relationship; According to the driving data mapping relationship, a driving data packet for the LCD screen display driving system is generated, and the driving data packet includes the final display brightness or voltage control value of each physical sub-pixel; The driving data packet is output to the LCD screen display driving system, and each physical sub-pixel of the LCD screen is driven to display according to the optimized virtual pixel image, the resolution-improved image content output is completed, and high-precision detail restoration and color restoration of the display content on the LCD screen are realized.

[0032] Embodiment 1: In order to verify the feasibility of the application in implementation, the application is applied to the liquid crystal screen resolution optimization scene of a large indoor advertising display terminal. In actual application, the terminal adopts a traditional LCD display panel, and the native resolution is 960x540, and the RGB vertical arrangement sub-pixel structure is adopted. Due to frequent replacement of advertising content and containing a large number of fine graphics and text, under the existing hardware conditions, the display effect often appears edge blur, color distortion and local detail loss, which is difficult to meet the needs of high-end brand advertisers for detail and color restoration.

[0033] In this scenario, first, the original low-resolution image data of the advertising material is collected by the content management system, and normalized and size-standardized processing is performed to ensure complete matching with the display screen size parameters. Subsequently, the system automatically identifies the physical parameters of the LCD screen, including resolution, actual physical screen size, and sub-pixel arrangement, and generates a virtual pixel mapping rule. Based on these parameters, an improved sub-pixel convolutional neural network is selected for end-to-end training of the normalized image data, and a high-resolution virtual pixel output image adapted to the current screen structure is obtained based on the virtual pixel mapping and multi-branch channel reorganization mechanism. In the optimization link, the system dynamically adjusts the sub-pixel weight according to the local content features of each frame of advertising image and the actual sub-pixel arrangement, and finally outputs the optimized virtual pixel image through the screen driving system to realize resolution enhancement without hardware modification.

[0034] In the actual continuous operation test process, standardized test materials and industry mainstream algorithms are used for comparative experiments. By comparing the display effects of different content types (such as brand logos, complex icons, and thin line texts), the clarity improvement rate, edge sharpness improvement, color uniformity, root mean square error (RMSE), and perceptual consistency indicators are recorded. Under the scheme of the present application, the average clarity of the processed image is improved by 18.3%, the edge sharpness is improved by 22.7%, and the color uniformity is improved by 15.6%. For the display error after resolution enhancement, the RMSE is reduced from 0.051 of the traditional convolutional network enhancement scheme to 0.032. The perceptual consistency is scored by multiple rounds of blind testing, and the average user score is improved from 3.8 to 4.6 (full score 5), with a high score evaluation rate of 86%. In the display of complex icons and line content, the detail restoration ability and boundary anti-aliasing ability of the display terminal of the present application are significantly improved.

[0035] Table 1 Comparison of LCD display performance of different methods

[0036] As can be seen from the above Table 1, the LCD screen resolution enhancement method based on virtual pixel mapping proposed by the present application has obvious advantages over the traditional convolutional super-resolution method and the common interpolation algorithm in multiple performance indicators. First, in terms of average clarity improvement, the present application scheme reaches 18.3%, which is higher than the traditional convolutional super-resolution of 10.1%, the bilinear interpolation of 6.3%, and the cubic interpolation of 7.9%, showing outstanding performance in detail texture restoration ability. Second, in terms of edge sharpness improvement, the present application reaches 22.7%, which is also superior to the traditional convolutional network of 12.4% and the two traditional interpolation methods of 7.1% and 8.2%, indicating that the present application can better restore the edge details of complex graphics and text in the image and improve the clarity of the visual boundary.

[0037] In terms of color uniformity, the improvement of the application is 15.6%, which is significantly higher than 7.8%, 3.6% and 4.1% of the comparative algorithms, indicating that the application can more effectively reduce color distortion and banding phenomenon with the help of virtual pixel recombination and adaptive sub-pixel fusion mechanism, and improve the overall color performance of the LCD screen. In terms of display error, the root mean square error (RMSE) is used for evaluation, and the application is only 0.032, which is further reduced compared with the traditional convolution super-resolution and interpolation algorithm, indicating that the numerical difference between the processed image and the high-resolution real image is smaller, and the reconstruction accuracy is higher.

[0038] In addition, the test results of subjective perception also fully verify the engineering effect of the application. The average score of users reaches 4.6, and the high score evaluation rate is 86%, which are both much higher than the traditional methods, showing that the application not only improves the objective display quality, but also greatly improves the subjective experience of the end user. In summary, the application scheme can realize all-round resolution improvement and visual optimization effect in the premise of not changing the LCD hardware, aiming at the detail restoration, edge sharpening, color uniformity and subjective perception of complex content, and has high practical application value

[0039] The above describes only the preferred specific embodiments of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and inventive concept of the application within the technical range disclosed by the application, which should be covered within the protection scope of the application.

Claims

1. A method for improving LCD screen resolution based on virtual pixel mapping, characterized in that, Includes the following steps: Acquire and preprocess low-resolution raw image data that needs to be displayed on the LCD screen to obtain a standardized low-resolution input image; Obtain the physical parameter information of the LCD screen and generate virtual pixel mapping rules; Based on the sub-pixel arrangement and virtual pixel mapping rules of the LCD screen, an improved sub-pixel convolutional neural network is constructed and trained to obtain an improved sub-pixel convolutional neural network model. The model inputs a standardized low-resolution image into an improved subpixel convolutional neural network and outputs a high-resolution feature image of the LCD screen. Based on the virtual pixel mapping rules, the high-resolution feature image of the LCD screen is recombined by multi-branch virtual pixel channels to obtain a virtual pixel recombined image; An adaptive subpixel fusion mechanism is adopted to dynamically adjust the subpixel weight allocation based on the content features of the virtual pixel reconstructed image and the subpixel arrangement of the LCD screen, thereby obtaining an optimized virtual pixel image. The display driver system optimizes the virtual pixel image and outputs it to the LCD screen according to the virtual pixel mapping rules, driving the LCD screen to display the image content with enhanced resolution.

2. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The standardized low-resolution input image is a multi-channel image data matrix that is normalized and size-standardized, with the same size as the LCD screen display area.

3. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The process of obtaining the physical parameter information of the LCD screen and generating virtual pixel mapping rules specifically includes: Obtain the resolution parameters of the LCD screen, which include the number of pixels in height and the number of pixels in width of the LCD screen display area; Obtain the physical size parameters of the LCD screen, and calculate the actual physical size and position distribution of each physical pixel and each sub-pixel in the display area based on the correspondence between the physical size and resolution of the LCD screen. Obtain the sub-pixel arrangement of the LCD screen and determine the arrangement order, type, and spatial position of each sub-pixel within each physical pixel; Based on the resolution parameters, physical size parameters, and sub-pixel arrangement of the LCD screen, a mapping relationship is established between each virtual pixel and the physical sub-pixels of the LCD screen, forming a virtual pixel mapping rule.

4. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The specific steps of constructing and training an improved sub-pixel convolutional neural network model based on the sub-pixel arrangement and virtual pixel mapping rules of the LCD screen include: An improved subpixel convolutional neural network model is constructed, which includes a subpixel-aware convolution module, a deep feature extraction module, a channel expansion convolution module, and a subpixel convolution rearrangement module. Based on the virtual pixel mapping rules, determine the initial parameters of the convolution kernel of the sub-pixel perceptual convolution module; The deep feature extraction module consists of multiple convolutional layers and stacked activation functions. It extracts high-dimensional feature representations from the input normalized image data and outputs a deep feature map. The channel-expanding convolutional module expands the number of channels in the deep feature map to obtain a channel-expanded feature map. The subpixel convolutional rearrangement module uses pixel rearrangement operations to spatially reorganize the channel-extended feature map into a high-resolution virtual pixel image; Using a standardized low-resolution input image as input and the corresponding high-resolution target image as the desired output, the improved sub-pixel convolutional neural network model is trained by defining a sub-pixel perception loss function for the sub-pixel arrangement of LCD screens. The improved subpixel convolutional neural network model is iteratively trained using the gradient descent method until the value of the subpixel perception loss function reaches the set convergence condition, and the trained improved subpixel convolutional neural network model is output.

5. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 4, characterized in that, The step of determining the initial parameters of the convolution kernel of the sub-pixel perceptual convolution module according to the virtual pixel mapping rule specifically includes: Based on the virtual pixel mapping rules, combined with the resolution, physical size and sub-pixel arrangement of the LCD screen, the spatial distribution characteristics of all types of sub-pixels in the display area are statistically analyzed and labeled to obtain the sub-pixel spatial index map; Based on the sub-pixel spatial index map, multiple sets of convolution kernels are designed for each seed pixel arrangement to obtain the convolution kernel size parameters and receptive field parameters; Based on the sub-pixel spatial index map, convolution kernel size parameters, and receptive field parameters, the number of convolution kernels in each group is set to be consistent with the number of sub-pixel structure types on the LCD screen to obtain the convolution kernel number parameters; Based on the virtual pixel mapping rules, the physical spatial position of the input virtual pixel in the LCD display area is mapped one by one with the corresponding physical sub-pixel unit to obtain the virtual pixel spatial mapping relationship; Based on the virtual pixel space mapping relationship, initialize the weight distribution of each convolution kernel to obtain the initial weight distribution parameters of the convolution kernel; For LCD screens with a non-linear sub-pixel arrangement, the receptive field and initial weight distribution of the convolution kernel are adjusted according to the virtual pixel space mapping relationship to obtain the parameters of the non-linear arrangement convolution kernel. Output parameters for the number of convolutional kernels, kernel size, receptive field, initial weight distribution, and non-linear arrangement of convolutional kernels.

6. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 4, characterized in that, The channel-expanded convolutional module expands the number of channels in the deep feature map to obtain the channel-expanded feature map, specifically including: Receive deep feature maps, count the original number of channels, spatial size and data format of the deep feature maps, and obtain the parameter information of the deep feature maps; Based on the LCD screen resolution improvement target, virtual pixel mapping rules, and spatial reorganization requirements of the subpixel convolutional rearrangement module, the required number of target output channels is calculated, and the target output channel number parameter is obtained. Based on the deep feature map parameter information and the target output channel number parameter, the number and size of the convolution kernels of the channel expansion convolution module are set to obtain the channel expansion convolution kernel configuration parameters; The deep feature map is input into the channel expansion convolution module. Using a preset convolution kernel configuration, all channels of the deep feature map are expanded and combined to output channel expansion feature map data.

7. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The process of inputting a standardized low-resolution input image into an improved sub-pixel convolutional neural network model and outputting a high-resolution feature image for the LCD screen specifically includes: The standardized low-resolution input image is fed into an improved subpixel convolutional neural network model trained according to the subpixel arrangement of the LCD screen and the virtual pixel mapping rules; In the sub-pixel perceptual convolution module, preliminary feature extraction is performed on the input image based on the sub-pixel arrangement characteristics of the LCD screen to obtain sub-pixel feature representations; In the deep feature extraction module, the sub-pixel feature representation is further processed by multi-layer convolution and high-dimensional feature extraction to obtain a deep feature map; In the channel-expanded convolution module, the deep feature map is expanded in the channel dimension according to the resolution improvement requirements to obtain the channel-expanded feature map; In the subpixel convolutional rearrangement module, the channel extended feature map is spatially reorganized, and the channel information is redistributed to the spatial dimension to output a high-resolution feature image of the LCD screen.

8. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The process of reconstructing a high-resolution feature image of an LCD screen using multi-branch virtual pixel channels based on virtual pixel mapping rules to obtain a reconstructed virtual pixel image specifically includes: Receive high-resolution feature images from the LCD screen to obtain high-resolution feature channel data; Based on the virtual pixel mapping rules, the correspondence between each channel of the high-resolution feature channel data and the physical sub-pixel type, arrangement order and spatial position of the LCD screen is analyzed to obtain the channel-sub-pixel mapping relationship; The high-resolution feature channel data is divided into multiple virtual pixel recombination branches according to the channel-sub-pixel mapping relationship to obtain virtual pixel recombination branch data; For each virtual pixel recombination branch, according to the virtual pixel mapping rules, the color components of the input virtual pixel are allocated to the corresponding physical sub-pixel units according to the spatial mapping and arrangement requirements to obtain the branch weighted recombination result; For different sub-pixel arrangements on LCD screens, the spatial combination order and weight ratio of each virtual pixel recombination branch are dynamically adjusted, and the weighted recombination results of all branches are integrated to form a virtual pixel recombination image.

9. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The adaptive sub-pixel fusion mechanism dynamically adjusts the sub-pixel weight allocation based on the content features of the virtual pixel reconstructed image and the sub-pixel arrangement of the LCD screen to obtain an optimized virtual pixel image. Specifically, this includes: Receive virtual pixel reconstructed images, analyze the spatial distribution, color composition, edge features, and texture distribution of the virtual pixel reconstructed images, and obtain image content feature data; Based on the sub-pixel arrangement of the LCD screen and combined with the virtual pixel mapping rules, an initial weight allocation scheme is set for each virtual pixel and its corresponding physical sub-pixel region to obtain the initial sub-pixel weight parameters. For the content feature data of virtual pixel reconstructed images, an adaptive weighting mechanism is adopted to dynamically adjust the weight parameters of sub-pixels within the spatial region to obtain adaptively adjusted weight parameters. By using a content-aware fusion algorithm, the initial sub-pixel weight parameters and the adaptively adjusted weight parameters are jointly weighted to calculate the final weight allocation of each color channel component of each virtual pixel on different physical sub-pixel units, thus obtaining the final sub-pixel weight allocation matrix. Based on the final sub-pixel weight allocation matrix, the virtual pixel reconstructed image is weighted and fused at each spatial location and color channel component to output an optimized virtual pixel image.

10. The method for improving LCD screen resolution based on virtual pixel mapping according to claim 1, characterized in that, The display driving system that outputs the optimized virtual pixel image to the LCD screen according to the virtual pixel mapping rule, and drives the LCD screen to display the image content with improved resolution, specifically includes: Receive the optimized virtual pixel image, and according to the virtual pixel mapping rules, match each virtual pixel and each color channel component in the optimized virtual pixel image to a specific sub-pixel unit in the physical display area of ​​the LCD screen to form a driving data mapping relationship; Based on the driving data mapping relationship, a driving data packet is generated, which includes the final display brightness or voltage control value of each physical sub-pixel; The driver data package is output to the LCD screen display driver system, which drives each physical sub-pixel of the LCD screen to display according to the optimized virtual pixel image.

Images