Virtual reality VR image recognition method and system based on Micro-LED

Abstract

The invention relates to the technical field of image enhancement, in particular to a Micro-LED-based virtual reality VR image recognition method and system, and the method comprises the following steps: based on an input image displayed by VR, analyzing the light intensity contribution and spatial arrangement mode of sub-pixels, namely, red, green and blue independent light-emitting units, of a Micro-LED display screen, and generating a sub-pixel spatial contribution distribution diagram; according to the method, the light intensity distribution of the sub-pixel space is analyzed, the light intensity contribution of each independent light-emitting unit to the image is extracted, the image correction is realized by combining the re-calculation of the pixel gray scale data, and the pixel-level gray scale reconstruction is realized on the basis, so that the image data correction precision is obviously improved; meanwhile, in the image distortion identification stage, the continuity of sub-block boundary gray difference is detected, and sensitivity sorting is performed on frequency energy distribution abnormal areas, so that the severity of compression distortion in the image is determined more accurately.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology, and in particular to a virtual reality (VR) image recognition method and system based on Micro-LED. Background Technology

[0002] Image enhancement technology is an important branch of digital image processing, mainly studying how to improve the recognizability of useful information, detail representation, and visual quality in images through specific algorithms or processing links. Virtual reality (VR) image recognition methods utilize the sub-pixel characteristics of Micro-LED display technology to address the compression distortion problem that occurs during VR image display. This involves analyzing the spatial light intensity contribution of sub-pixels and combining this with the visual perception characteristics of the human eye to perform region-based image processing, thereby improving image detail and edge sharpness in VR scenes.

[0003] Current technologies fail to adequately distinguish between the spatial contribution characteristics of subpixels and the differences in human visual perception, relying too heavily on holistic image enhancement strategies. This results in insufficient local optimization of image details and sensitive areas. Furthermore, existing technologies only analyze compression distortion problems on a holistic or large scale, making it difficult to locate and process local frequency anomalous regions. This can easily lead to residual compression distortion details, thus affecting the overall visual experience of the image. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose a virtual reality (VR) image recognition method and system based on Micro-LED.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a virtual reality (VR) image recognition method based on Micro-LED, comprising the following steps: Based on the input image of the VR display, the light intensity contribution and spatial arrangement of the sub-pixels of the Micro-LED display, namely the independent light-emitting units of red, green and blue, are analyzed to generate a sub-pixel spatial contribution distribution map. The pixel data is recalculated by combining the original image data with the contribution value of each independent light-emitting unit in the sub-pixel spatial contribution distribution map to obtain sub-pixel corrected image data. Based on the subpixel-corrected image data, the presence of block boundary features is detected, anomalies in frequency energy distribution are analyzed, the severity of compression distortion is determined, compression distortion detail parameters are obtained, and based on the compression distortion detail parameters, differential processing coefficients are set for different regions of the image to obtain a differential processing weight set. Based on the subpixel-corrected image data and the differential processing weight set, the parameters of the deblocking filter are adjusted by calling the region processing coefficients in the differential processing weight set and applied to generate an anti-compression distortion corrected image. Based on the anti-compression distortion corrected image and the subpixel spatial contribution distribution map, the human eye's perception threshold for information under different brightness and contrast values ​​in the anti-compression distortion corrected image is calculated. Based on the perception threshold and referring to the weight of each independent luminous unit in the subpixel spatial contribution distribution map, sharpening adjustment is performed to obtain visual salient feature data. The visual salient feature data is then fused with the anti-compression distortion corrected image to generate a visual recognition optimized image.

[0006] Preferably, the steps for obtaining the sub-pixel spatial contribution distribution map are as follows: Based on the input image of the VR display, the coordinate position of each pixel is located in the image rendering matrix. The geometric centroid coordinates, luminous area and brightness value of the corresponding red light-emitting unit on the Micro-LED display are matched pixel by pixel. The Euclidean distance from the centroid of each light-emitting unit to the center of the target pixel is measured to form a set of spatial brightness parameters of the red light-emitting unit. Based on the set of spatial brightness parameters of the red light-emitting unit, calculate the relative light intensity contribution factor of the red light-emitting unit to the pixel; Based on the relative light intensity contribution factor, the green and blue light-emitting units are also calculated using the relative light intensity contribution factor formula. Channel fusion and pixel coordinate reverse mapping are performed on the three-color contribution factor of each pixel to generate a sub-pixel spatial contribution distribution map.

[0007] Preferably, the step of acquiring the subpixel-corrected image data is as follows: Based on the original image data, each pixel of the original image data is scanned pixel by pixel, the initial gray values ​​of the red, green and blue channels corresponding to the pixel are extracted, and the initial gray value matrix of the three channels of each pixel is established to obtain the original pixel gray value matrix. Based on the original pixel grayscale matrix, according to the sub-pixel spatial contribution distribution map, and referring to the relative light intensity contribution factors of the three independent light-emitting units of red, green and blue at each pixel position, the product of the original grayscale value and the corresponding channel relative light intensity contribution factor is calculated for each channel, and the product results of the three channels are normalized to form the sub-pixel channel grayscale correction matrix. Based on the subpixel channel grayscale correction matrix, the corrected grayscale values ​​of the red, green, and blue channels are recombined pixel by pixel, mapped back to the image coordinate system point by point, the image three-channel grayscale correction data is reconstructed, and the image color channels are overlapped and fused to obtain subpixel corrected image data.

[0008] Preferably, the step of obtaining the compression distortion details parameters is as follows: Based on the subpixel-corrected image data, the image is divided into multiple sub-blocks of fixed size. The boundary pixel column between each sub-block and its adjacent sub-blocks on the right and below is traversed. The absolute value sequence of the difference between the gray values ​​of corresponding pixels on both sides of the boundary line is calculated, and the absolute value sequence is averaged by a sliding window. If the average gray value difference in three consecutive windows is greater than twice the average gray value difference of the entire image, the sub-block is marked as a sub-block with significant boundary differences, and a boundary difference label list is generated. Based on the boundary difference marker list, calculate the frequency energy anomaly value for each marked sub-block; Based on the frequency energy anomalies, all sub-blocks whose frequency energy anomalies are greater than the sum of the mean and standard deviation of the frequency energy anomalies of the entire sub-block are set as the compression distortion core sub-block set. The horizontal and vertical grayscale gradient amplitudes of all pixels in each sub-block of the compression distortion core sub-block set are extracted, and the pixel gradients are classified according to the principal gradient direction to obtain the compression distortion detailed parameters.

[0009] Preferably, the step of obtaining the differentiated processing weight set is as follows: Based on the compression distortion details parameters, the pixel position coordinates and corresponding pixel gradient magnitude values ​​of all sub-blocks in the compression distortion core sub-block set are extracted. The pixel gradient magnitude values ​​of each sub-block are sorted from largest to smallest. The maximum and average gradient magnitude values ​​after sorting are recorded. Sub-blocks whose maximum pixel gradient magnitude value exceeds twice the average gradient are marked as high distortion sensitive regions, and a high distortion sensitive region marker list is generated. Based on the high-distortion sensitive area marker list, the viewpoint tracking data provided by VR is called to determine the coordinate position of the center area of ​​the field of view where the user's viewpoint is focused. The spatial distance from each sub-block in the high-distortion sensitive area to the center area of ​​the field of view is calculated based on the center area of ​​the field of view. The high-distortion sensitive areas are sorted according to the spatial distance to form a region priority sorting list. Based on the region priority sorting list, a preset processing intensity level is matched for each sub-block in the sorting list to generate a differentiated processing weight set.

[0010] Preferably, the step of acquiring the compression-resistant image is as follows: Based on the subpixel corrected image data, the corrected grayscale values ​​of the red, green, and blue channels of all pixels in each sub-block are extracted pixel by pixel. The mean, variance, and peak position of the grayscale distribution of each channel of each sub-block are statistically analyzed to generate a grayscale statistical feature set of the sub-block. Based on the gray-scale statistical feature set of the sub-blocks, the differential processing weight set is called to match the regional processing intensity of each sub-block, and the filtering radius and filtering intensity of the deblocking filter corresponding to each sub-block are adjusted one by one based on the regional processing intensity to form a differential filtering parameter set for each sub-block deblocking filter. Based on the differentiated filtering parameter set, the corresponding filtering parameters are called for each sub-block, and the sub-pixel correction grayscale values ​​of each sub-block are filtered by convolution operation. The filtered sub-block image data are then smoothly stitched together to reconstruct all sub-block images and map them back to the original image space coordinates, generating an anti-compression distortion correction image.

[0011] Preferably, the steps for obtaining the visual salient feature data are as follows: Based on the anti-compression distortion corrected image, the gray values ​​of the red, green and blue channels of each pixel in the image are extracted pixel by pixel. The brightness value and local area contrast value corresponding to the gray values ​​of all pixels are calculated respectively. The brightness value and local area contrast value are segmented and classified respectively. The number of pixels and the average gray value of each brightness segment and contrast segment are counted to generate the brightness and contrast segment statistical features. Based on the segmented statistical characteristics of brightness and contrast, the relative light intensity contribution factor of each independent light-emitting unit in the sub-pixel spatial contribution distribution map is called to determine the human eye's perception threshold of image information under each brightness and contrast segment. The perceptual threshold is used as a benchmark to determine the adjustment range of pixel grayscale sharpening segment by segment, forming a pixel grayscale sharpening adjustment range mapping relationship. Based on the pixel grayscale sharpening adjustment amplitude mapping relationship, the grayscale value of the corresponding pixel in the anti-compression distortion corrected image is adjusted pixel by pixel, and the local contrast features and edge saliency features of the image pixels after grayscale sharpening adjustment are extracted to obtain visual saliency feature data.

[0012] Preferably, the step of acquiring the visual recognition optimized image is as follows: Based on the aforementioned visual saliency feature data, local contrast features and edge saliency features of each pixel are extracted pixel by pixel. The feature intensity value of each pixel and the position coordinates of the corresponding pixel in the image coordinate space are recorded. The high, medium and low saliency regions of the pixels are marked according to the magnitude of the feature intensity value to generate a visual saliency marking map. Based on the visual saliency marker map, the gray values ​​of the red, green, and blue channels of the corresponding positions in the anti-compression distortion corrected image are called. Texture enhancement is performed on the gray values ​​of pixels in high saliency areas, edge enhancement is performed on the gray values ​​of pixels in medium saliency areas, and noise suppression is performed on the gray values ​​of pixels in low saliency areas, thus forming regional gray-scale adjustment image data. Based on the grayscale adjusted image data, the high, medium, and low saliency regions after grayscale adjustment are smoothly and gradually merged pixel by pixel, and the grayscale boundaries of each channel of the merged image are smoothly interpolated to reconstruct continuous and unified color image data, thus obtaining a visual recognition optimized image.

[0013] This invention provides a virtual reality (VR) image recognition system, comprising: The sub-pixel analysis module analyzes the light intensity contribution and spatial arrangement of the sub-pixels of the Micro-LED display screen, i.e., the light-emitting units of red, green and blue independent light-emitting units, based on the input image of the VR display. It generates a sub-pixel spatial contribution distribution map, and recalculates the pixel data by combining the original image data with the contribution value of each independent light-emitting unit in the sub-pixel spatial contribution distribution map to obtain sub-pixel corrected image data. The compression distortion analysis module, based on the subpixel corrected image data, detects the presence of block boundary features, analyzes anomalies in frequency energy distribution, determines the severity of compression distortion, obtains compression distortion detail parameters, and sets differential processing coefficients for different regions of the image based on the compression distortion detail parameters to obtain a differential processing weight set; The deblocking filter module, based on the subpixel corrected image data and the differential processing weight set, calls the region processing coefficients in the differential processing weight set to adjust the parameters of the deblocking filter and applies them to generate an anti-compression distortion corrected image. The visual perception enhancement module calculates the human eye's perception threshold for information at different brightness and contrast values ​​in the anti-compression distortion corrected image based on the anti-compression distortion corrected image and the sub-pixel spatial contribution distribution map. Based on the perception threshold and referring to the weight of each independent luminous unit in the sub-pixel spatial contribution distribution map, it performs sharpening adjustment to obtain visual salient feature data. The visual salient feature data is then fused with the anti-compression distortion corrected image to generate a visual recognition optimization image.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention analyzes the sub-pixel spatial light intensity distribution, extracts the light intensity contribution of each independent emitting unit to the image, and combines it with recalculated pixel grayscale data to achieve image correction. Furthermore, it achieves pixel-level grayscale reconstruction, significantly improving the accuracy of image data correction. Simultaneously, in the image distortion recognition stage, it detects the continuity of grayscale differences at sub-block boundaries and ranks areas with abnormal frequency energy distribution by sensitivity, thereby more accurately determining the severity of compression distortion in the image. In addition, it integrates viewpoint tracking data as a reference for the field of view center, implementing differentiated processing intensity adjustments based on the visual sensitivity of different regions in the image to ensure more refined image quality in visually focused areas. Furthermore, based on the human eye perception threshold, it combines differentiated sharpening processing with sub-pixel spatial light intensity contribution factors to further enhance visual perception and image detail clarity, effectively improving the sharpness and realism of image details and edge areas, enhancing the visual expressiveness and immersive experience of VR images. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] Please see Figure 1 This invention provides a technical solution: a virtual reality (VR) image recognition method based on Micro-LED, comprising the following steps: Based on the input image of the VR display, the light intensity contribution and spatial arrangement of the sub-pixels of the Micro-LED display, namely the independent light-emitting units of red, green and blue, are analyzed to generate a sub-pixel spatial contribution distribution map. The pixel data is recalculated by combining the original image data with the contribution value of each independent light-emitting unit in the sub-pixel spatial contribution distribution map to obtain sub-pixel corrected image data. Based on subpixel-corrected image data, the presence of block boundary features is detected, anomalies in frequency energy distribution are analyzed, the severity of compression distortion is determined, and compression distortion detail parameters are obtained. Based on the compression distortion detail parameters, differential processing coefficients are set for different regions of the image to obtain a differential processing weight set. Based on subpixel-corrected image data and a differential processing weight set, the parameters of the deblocking filter are adjusted by calling the region processing coefficients in the differential processing weight set and then applied to generate an anti-compression distortion corrected image. Based on the anti-compression distortion corrected image and the subpixel spatial contribution distribution map, the human eye's perception threshold for information under different brightness and contrast values ​​in the anti-compression distortion corrected image is calculated. Based on the perception threshold and referring to the weight of each independent luminous unit in the subpixel spatial contribution distribution map, sharpening adjustment is performed to obtain visual salient feature data. The visual salient feature data is then fused with the anti-compression distortion corrected image to generate a visual recognition optimized image.

[0018] The steps to obtain the sub-pixel spatial contribution distribution map are as follows: Based on the input image of the VR display, the coordinate position of each pixel is located in the image rendering matrix. The geometric centroid coordinates, luminous area and brightness value of the corresponding red light-emitting unit on the Micro-LED display are matched pixel by pixel. The Euclidean distance from the centroid of each light-emitting unit to the center of the target pixel is measured to form a set of spatial brightness parameters of the red light-emitting unit. Based on the spatial brightness parameter set of the red emitting unit, the relative light intensity contribution factor of the red emitting unit to the pixel is calculated using the following formula: ; in, The relative light intensity contribution factor for red emitting units. For the first The brightness value of each red light-emitting unit (cd / m²) 2 ), For the first The luminous area of ​​each red luminous unit (m²) 2 ), From the center of the pixel to the 1st The Euclidean distance (m) between the centroids of the red luminous units. It is the Gaussian distance decay function (dimensionless). The spatial attenuation characteristic distance constant (m) is given. For the first The effective light intensity (cd) of each light-emitting unit in the current frame. The correction constant (cd) is the sum of the squares of the light intensity. 2 ), This represents the number of red light-emitting units corresponding to the current pixel. Based on the relative light intensity contribution factor, the green and blue light-emitting units are also calculated using the relative light intensity contribution factor formula. Channel fusion and pixel coordinate inverse mapping are performed on the three-color contribution factors of each pixel to generate a sub-pixel spatial contribution distribution map.

[0019] Specifically, for input images used in VR displays, the system first preprocesses the image, performing color space correction and noise filtering to ensure that the image data is consistent with the data format requirements of subsequent processing modules. Next, the system iterates through each pixel in the rendering matrix (i.e., frame buffer) output by the image rendering pipeline and reads its two-dimensional screen coordinates rendered in the virtual scene. For each pixel traversed, the system queries a pre-built and stored lookup table of physical parameters for the Micro-LED display. This lookup table records the physical characteristics of each individual light-emitting unit (red R, green G, blue B subpixels) that makes up the display panel, including its precise geometric center coordinates on the display substrate. (Unit: micrometers), Design effective luminescent area (Unit: square micrometers), and calibrated brightness value at a standard test drive current (e.g., 0.1 microamps). (Unit: nits) For the target pixel being processed, taking its red (R) channel as an example, the system will search within its defined neighborhood (e.g., around the center of the target pixel) within its defined neighborhood search area. With the geometric center as the center, the search radius is... (The circular area) retrieves all red Micro-LED light-emitting units. The value is not fixed, but dynamically adjusted or preset according to the sub-pixel density and arrangement of Micro-LEDs (such as PenTile, RGBStripe, etc.). The setting principle is to ensure that all neighboring sub-pixels of the same color that make a significant contribution to the light intensity of the current pixel can be covered, while avoiding the introduction of too many distant sub-pixels with weak contributions to reduce computational complexity. For example, for a display with a sub-pixel pitch of 20 micrometers, It can be set to 30 micrometers, and this value is determined through offline calibration: selecting different... Values ​​(e.g., 20µm, 25µm, 30µm, 35µm, 40µm) are used to calculate the color reproduction accuracy and perceived sharpness scores of the final image. A balance point is then selected based on the computation time, such as when... Increasing the pixel size from 30µm to 35µm improves color accuracy by less than 0.5%, but increases computation time by 20%. Therefore, 30µm is considered the optimal value. For each red emitting unit found within this search region (index 1...),... Record its calibrated geometric centroid coordinates. Its actual luminescent area was determined through high-precision microscopic imaging and image recognition algorithms. And the actual brightness value of the light-emitting unit in the current display frame, obtained in real time through feedback from the driving circuit or real-time monitoring by a photoelectric sensor. Then, the centroid of each matched red emitting unit is calculated. To the center of the current target pixel The linear geometric distance, i.e., the Euclidean distance. This distance is obtained by taking the square root of the sum of the squares of the coordinate differences on the corresponding coordinate axes. The above series of parameters (including the unit index) for each red emitting unit are then used. centroid coordinates luminous area Real-time brightness and the calculated Euclidean distance The data entries of all these red luminescent units are organized into a structured data entry. For the current target pixel, the data entries of all these red luminescent units are collected together to form the spatial brightness parameter set of the red luminescent units.

[0020] formula: The Gaussian distance decay function is: The advantage of this formula lies in its ability to simulate the combined visual contribution of several sub-pixel light-emitting units of the same color adjacent to a specific pixel in a Micro-LED display. It integrates the actual emitted brightness of each sub-pixel unit through a weighted summation of the molecular components. ), physical luminescent area ( And its spatial position relative to the pixel center (via the Gaussian distance decay function) and its influence. This reflects the principle that the closer the distance, the greater the impact, which aligns with the characteristics of human vision, thus yielding a total effective luminous flux contribution. The denominator is based on the effective luminous intensity of each sub-pixel unit (…). The squares are summed and the square root is taken, similar to a calculation of an energy or intensity norm, and then a small correction constant is added. This design ensures the numerical stability of the calculations, serving a normalization and balancing function. This results in a final relative light intensity contribution factor... It can more realistically reflect the actual contribution of subpixels, rather than simply averaging or only considering the nearest neighbor unit, providing more refined and physically meaningful weight parameters for subsequent subpixel-level image correction, thus improving the realism and clarity of virtual reality images displayed on Micro-LED screens.

[0021] The steps to obtain each parameter are as follows: For the first Brightness value of each red light-emitting unit (unit: cd / m²) 2 This parameter is obtained through calibration of the Micro-LED display panel. In a controlled darkroom environment, using a luminance meter, the measured value on the display screen is... A specific driving current is applied to each red light-emitting unit (e.g., a 0.2µA driving current precisely output to the target subpixel via an FPGA-controlled driving board). After the light emission stabilizes, multiple (e.g., 5) brightness readings are taken, and the average value is used as the brightness value of that unit under that driving condition. This process requires measuring multiple sampling points on the panel (e.g., representative red subpixels in the center, corners, and edges of the screen) to obtain statistically representative brightness data or establish a model of the relationship between brightness and driving current. For example, for a red subpixel in the center area of ​​the screen, under a 0.2µA driving current, the 5 measurements are 1250.5, 1251.0, 1250.8, 1251.2, and 1250.0 cd / m², respectively. 2 Its average brightness The calculation is as follows: ; For the first The luminous area of ​​each red luminous unit (unit: m²) 2The Micro-LED display sample is placed on a microscope stage, and a single red emitting unit is observed using, for example, a 100x objective lens. A microscopic image of the emitting unit in its luminescent state is captured and imported into image analysis software. Using edge detection algorithms and threshold segmentation techniques, the contour of the luminescent region is extracted. The software automatically calculates the area enclosed by the contour based on a pre-calibrated pixel size (µm / pixel). For example, if the microscope system is calibrated to show each image pixel representing an actual physical size of 0.05µm x 0.05µm, and the software analysis indicates that a certain red emitting unit occupies 1600 pixels, then its luminescent area... The calculation is as follows: ; From the center of the pixel to the 1st The Euclidean distance (in meters) of the centroids of each red emitting unit: This parameter is calculated based on the sub-pixel layout design data during display manufacturing and the mapped position of the currently rendered pixel on the screen. First, the physical centroid coordinates of each red emitting unit are extracted from the display's design specifications or GDSII layout file. Meanwhile, the physical coordinates of the center of the target pixel whose contribution is to be calculated on the display screen are: This mapping is determined by the rendering pipeline of the VR system. Then the Euclidean distance... The calculation formula is: For example, the physical coordinates of the target pixel center are... A red light-emitting unit nearby The coordinates of the centroid are Then the distance between them The calculation is as follows: ; Spatial attenuation characteristic distance constant (unit: m): This parameter is a key parameter in the Gaussian distance attenuation function, determining the rate at which subpixel contribution decays with distance. Its value needs to be set with reference to the subpixel spacing of the display and the point spread function characteristics of the human visual system. A common setting method is to set it to a value that represents the average subpixel spacing of the same color. times, The value of is typically between 0.5 and 2, and needs to be optimized through visual experiments. For example, for an average red subpixel spacing of , The display screen can test a series of Value, such as By generating a series of standard test images (such as checkerboard patterns and line patterns), under different... The subpixel contribution is calculated and the image is rendered under a set value. Then, multiple observers subjectively rate the image's sharpness, ringing effect, artifacts, etc. (e.g., using the MOS score, 1-5 scale). Objective metrics such as MTF (modulation transfer function) can also be used for evaluation. The method that achieves the best balance between subjective and objective evaluation is selected. Value. For example, after testing, when At this time, the overall visual effect of the image is at its best.

[0022] For the first Effective light intensity (in cd) of a single luminous unit in the current frame: This parameter characterizes the light intensity emitted by a single subpixel unit in a specific direction (usually the normal direction). It can be derived from its brightness. and luminous area Approximate calculations can be performed. For example, the normal light intensity of a single subpixel under a specific drive current can be directly measured using a specialized light intensity probe. Alternatively, if the electro-optical conversion efficiency of the device is known... (cd / W) and operating voltage Current ,but It can be done Calculate the electrical power and then multiply it by the efficiency. An approximate calculation is used here: if... and ,but .

[0023] The correction constant is the sum of the squares of the light intensity (unit: cd). 2 ): This is a regularization term introduced to prevent the denominator from being equal to zero in extreme cases.

[0024] Number of red emitting units corresponding to the current pixel: This parameter refers to the total number of red emitting units actually involved in the calculation within the neighborhood defined for calculating the red channel contribution of the current pixel. For example, within a search radius, if coordinate matching and filtering ultimately determine that the contribution of 3 red emitting units needs to be included, then... .

[0025] Calculation process: For example, for a specific pixel, within its defined neighborhood, there are... The individual red emitting units contribute significantly to this. The relevant parameter values ​​are set (or obtained using the aforementioned method) as follows: For the first red light-emitting unit ( ): ; ; ; ; For the second red light-emitting unit ( ): (For example, this unit is slightly less bright); (For example, this unit has a slightly larger area); (For example, this unit is a little far away); ; Shared parameters: ; ; Step 1: Calculate the Gaussian distance decay function ; ; ; Step 2: Calculate the numerator ; Item 1 ( ): ; Item 2 ( ): ; Sum of numerators: ; Step 3: Calculate the denominator ; ; ; Denominator: (because (Very small, its impact is negligible at this level of precision); Step 4: Calculate the relative light intensity contribution factor ; ; This result indicates that, for the pixel under investigation, approximately 94.721% of the relative contribution to the perceived brightness of its red channel comes from the combined effect of the two neighboring red emitting units. This value will be directly used in subsequent steps to adjust the grayscale value of the red channel in the original image data, simulating the impact of Micro-LED subpixel layout and luminescence characteristics on the final pixel color. A higher value (such as 0.94721 in this example) means that the contribution of these subpixels to the red color of this pixel is very direct and dominant.

[0026] The relative light intensity contribution factor of the red emitting unit to the target pixel, calculated in the previous step. The system will retrieve the spatial brightness parameter sets of the green emitting unit (G) and the blue emitting unit (B) for the same target pixel, and fully adopt the calculation... The formula for calculating the relative light intensity contribution factor and its parameters (such as the spatial attenuation characteristic distance constant) are the same. Correction constant for summation of squared light intensity The relative light intensity contribution factor of each green emitting unit was calculated independently. The relative light intensity contribution factor of blue emitting units This means that for the green channel, the formula in... , , , and Replace with the brightness value of the corresponding green subpixel. luminous area ,distance Effective light intensity and quantity Similarly, for the blue channel, the red, green, and blue relative light intensity contribution factors for a single target pixel are calculated. After the initial calculation, the system extends this process to every pixel in the VR input image, generating a set of three-dimensional contribution factor vectors for each pixel. Next, channel fusion and inverse pixel coordinate mapping are performed. Specifically, channel fusion refers to... The values ​​are organized into a two-dimensional map (R-Map) representing the contribution of the red channel, with all pixels... The values ​​are organized into a two-dimensional map (G-Map) representing the contribution of the green channel, showing the values ​​of all pixels. The values ​​are organized into a two-dimensional map (B-Map) representing the contribution of the blue channel. The dimensions (width and height) of these three maps are completely consistent with the dimensions of the original VR input image. The inverse mapping of pixel coordinates ensures that the coordinates in the R-Map are identical. place The value corresponds precisely to the coordinates in the original image. The red subpixel contribution of the pixel at that location is calculated similarly for G-Map and B-Map, so that each pixel has a triplet. The set of these three two-dimensional maps (R-Map, G-Map, B-Map) together constitutes the final sub-pixel spatial contribution distribution map.

[0027] The steps for acquiring subpixel corrected image data are as follows: Based on the original image data, each pixel of the original image data is scanned pixel by pixel, the initial gray values ​​of the red, green and blue channels corresponding to the pixel are extracted, and the initial gray value matrix of the three channels of each pixel is established to obtain the original pixel gray value matrix. Based on the original pixel grayscale matrix, according to the sub-pixel spatial contribution distribution map, and referring to the relative light intensity contribution factors of the three independent light-emitting units of red, green and blue at each pixel position, the product of the original grayscale value and the corresponding channel relative light intensity contribution factor is calculated for each channel, and the product results of the three channels are normalized to form the sub-pixel channel grayscale correction matrix. Based on the subpixel channel grayscale correction matrix, the corrected grayscale values ​​of the red, green and blue channels are recombined pixel by pixel, mapped back to the image coordinate system point by point, the three-channel grayscale correction data of the image is reconstructed, and the image color channels are overlapped and fused to obtain subpixel corrected image data.

[0028] Specifically, based on the original image data, the scan is first performed in a predetermined order, for example, starting from the first row and first column of pixels in the image, scanning row by row and column by column within a row, until all pixels in the image have been processed. Then, each pixel is scanned... At that time, among them Column coordinates representing pixels, Representing the row coordinates of a pixel, the system directly extracts the initial grayscale values ​​of its corresponding red (R), green (G), and blue (B) color channels from the pixel's original data structure. For example, for an image stored in 24-bit RGB format, each color channel of each pixel is typically represented by an 8-bit unsigned integer, with a value ranging from 0 to 255. These three extracted values... , , This refers to the original brightness or intensity information of the pixel in each color channel. Subsequently, the system will perform the same extraction operation for every pixel in the image and organize and store these extracted initial grayscale values ​​of the three channels in an orderly manner according to their spatial position relationship in the original image. Specifically, three independent two-dimensional arrays can be constructed, corresponding to the R, G, and B channels respectively. The dimensions of the arrays are consistent with the height and width of the original image. Each array element stores the grayscale value of the corresponding pixel in that channel. Alternatively, more commonly, a three-dimensional array can be constructed, whose dimensions are usually [image height × image width × 3]. The third dimension index 0, 1, and 2 respectively store the initial grayscale values ​​of the R, G, and B channels. This structured data set that completely contains the initial grayscale information of the three channels of all pixels is the original pixel grayscale matrix.

[0029] Based on the original pixel grayscale matrix obtained in the previous step, and by calling the previously generated sub-pixel spatial contribution distribution map, the system begins to calculate the corrected grayscale value for each color channel of each pixel. The specific operation is as follows: For any pixel in the image... First, the initial grayscale value of the red channel is obtained from the original pixel grayscale matrix. Simultaneously, the contribution distribution map of the sub-pixel space is queried and the pixel is obtained. The relative light intensity contribution factor of the perfectly corresponding red emitting units This factor, calculated in a previous step, reflects the actual light intensity contribution of surrounding red sub-pixels to the red channel of the logical pixel. Next, the product of these two factors is calculated to obtain the preliminary red channel correction value. The exact same operation is performed on both the green and blue channels, i.e., obtaining the relevant data respectively. and and the corresponding relative light intensity contribution factor. and Calculate the preliminary green channel correction value. and blue channel correction value After completing the product calculation for all channels of all pixels, since the numerical range of the product result may no longer be the standard display grayscale range (e.g., 0-255), it is necessary to normalize all product results for each color channel separately. Taking the red channel as an example, the system will traverse all pixels in the image. Find the minimum value among the values. and maximum value Then for each pixel The grayscale value is adjusted using a linear normalization formula, mapping it to a preset display grayscale range, such as the standard 8-bit grayscale range of 0 to 255. The normalization calculation formula is as follows: ,in This indicates rounding to the nearest integer. equal (That is, all product values ​​within the channel are the same), then all Set it directly to the product value (if the value is within 0-255) or the median of the target grayscale range of the channel (e.g., 128), for the green channel. Values ​​and the blue channel The values ​​also undergo the exact same normalization process independently, yielding the following results: and After this step, each color channel of each pixel has a normalized corrected grayscale value, and these values ​​together constitute the sub-pixel channel grayscale correction matrix.

[0030] Based on the sub-pixel channel grayscale correction matrix generated in the previous step, this matrix contains the red values ​​of each pixel in the image after weighting and normalization by the sub-pixel contribution factor. ,green and blue The system then performs data reconstruction and image reconstruction pixel by pixel, based on the corrected grayscale values ​​of the three channels. The system extracts the three corrected grayscale values ​​corresponding to each sub-pixel channel from the grayscale correction matrix: , and The system then recombines these three values ​​into a basic unit representing the color information of that pixel. For example, in a standard RGB image format, these three values ​​will be arranged in a specific order (such as R, G, B) to form a pixel's color vector. Next, the system accurately maps these recombined pixel color information point-by-point back to their corresponding spatial positions in the original image coordinate system. This process is equivalent to reconstructing a complete image data structure containing all corrected pixel color data based on the dimensions (height and width) and pixel arrangement order of the original image. This structure is the reconstructed three-channel grayscale correction data of the image. The last step is to perform image color channel overlap and fusion. This step is essentially to conceptually or practically merge the data of the three corrected color channels (red, green, and blue) that were calculated and stored independently for each pixel into a multi-channel color image, so that the visual information of the three channels can be correctly superimposed and mixed during display or subsequent processing to present the final color effect. Through this series of operations, subpixel-corrected image data with subpixel level correction is obtained.

[0031] The steps to obtain compression distortion details parameters are as follows: Based on subpixel-corrected image data, the image is divided into multiple sub-blocks of fixed size. The boundary pixel column between each sub-block and its adjacent sub-blocks on the right and below is traversed. The absolute value sequence of the difference between the gray values ​​of corresponding pixels on both sides of the boundary line is calculated, and the absolute value sequence is averaged by a sliding window. If the average gray value difference in three consecutive windows is greater than twice the average gray value difference of the entire image, the sub-block is marked as a sub-block with significant boundary difference, and a boundary difference label list is generated. Based on the boundary difference marker list, the frequency energy anomaly value of each marked sub-block is calculated using the following formula: ; in, For the first Frequency energy anomalies of individual sub-blocks For sub-blocks The Middle The index of a pixel, For sub-blocks Total number of pixels For the first grayscale value of each pixel. This represents the grayscale gradient of the pixel in the horizontal direction. This represents the grayscale gradient of the pixel in the vertical direction. This represents the local average gray level of the pixel's neighborhood. To prevent constants with a denominator of zero; Based on frequency energy anomalies, all sub-blocks with frequency energy anomalies greater than the sum of the mean and standard deviation of the frequency energy anomalies of all sub-blocks in the image are set as the core sub-block set of compression distortion. The horizontal and vertical grayscale gradient amplitudes of all pixels in each sub-block of the core sub-block set of compression distortion are extracted. The pixel gradients are classified according to the principal gradient direction to obtain the compression distortion details parameters.

[0032] Specifically, based on the subpixel-corrected image data obtained from the aforementioned steps, the image data (if it is a color image, it is first converted to a grayscale image, for example, by taking the average of the R, G, and B channels, or by using a standard brightness calculation formula) is first processed. Get the grayscale value of each pixel Logically, the image is divided into a series of non-overlapping, fixed-size sub-blocks. For example, the sub-block size can be set to 8×8 pixels or 16×16 pixels. This size is chosen based on the image resolution and the expected scale of the compression block effect. For instance, for a 1920×1080 resolution image, using 16×16 pixel sub-blocks would yield the following results. This is a sub-block array of 120 rows and 68 columns (if the last column is insufficient, it will be padded to the actual width). The system then iterates through each sub-block in the image (excluding the rightmost and bottommost sub-blocks) and checks its boundaries with its right-hand adjacent sub-block and its bottom adjacent sub-block. Taking the right boundary as an example, the system extracts the grayscale value of the rightmost column of pixels in the current sub-block and the grayscale value of the leftmost column of pixels in its right-hand adjacent sub-block. It calculates the absolute value of the difference between the grayscale values ​​of corresponding pixels in these two columns (i.e., pixel pairs in the same row), forming a sequence containing... One difference ( The system extracts the absolute value sequence of the difference between the grayscale values ​​of the corresponding pixels at the bottom row of the current sub-block and the top row of the adjacent sub-block below. Then, it applies a sliding window averaging smoothing process to these generated absolute value sequences. The sliding window size is set, for example, a window length of 5 pixels and a sliding step of 1 pixel. The arithmetic mean of the absolute differences within each window is calculated. Before this operation, the system first calculates the overall average of the absolute values ​​of the grayscale differences of all corresponding pixels at all such boundaries (including the right and bottom boundaries of all sub-blocks) in the entire sub-pixel corrected image data, as the "average grayscale difference of the entire image". This is obtained by summing the absolute values ​​of all boundary differences and dividing by the total number of differences. For example, if a total of 10,000 absolute values ​​of boundary differences are calculated for the entire map, their sum is 50,000. Then, for each boundary of each sub-block (the right and bottom boundaries are judged separately), if there are three consecutive windows with average grayscale differences greater than the threshold in the difference sequence after the sliding window averaging process... ,in Set it to twice the "average grayscale difference of the entire image", that is... If the current sub-block is initially determined to have a significant difference on the corresponding boundary, the sub-block is marked as a sub-block with a significant boundary difference once any boundary (right or bottom) of a sub-block meets this condition. The identifiers of all such sub-blocks (e.g., their index or starting coordinates in the sub-block array) are summarized to generate a list of boundary difference markers.

[0033] formula: The advantage of this formula lies in its ability to effectively quantify the degree of frequency energy anomalies within image sub-blocks, particularly for artifacts caused by compression distortion (such as block artifacts). Its design comprehensively considers the horizontal gradient, vertical gradient, and local average gray level of pixels. The molecular part is determined through the horizontal gradient... With a vertical gradient squared and local average gray level The product of logarithmic terms is used to capture complex texture and edge information in the neighborhood of a pixel. It is more sensitive to strong gradients (especially in the vertical direction), and The introduction of this technology makes it possible to achieve smoothness in the region ( Smaller gradient changes are more pronounced, while in areas with complex textures ( Larger values ​​(increased values) are relatively suppressed, and this combination can better reflect the block boundary discontinuities or internal texture loss caused by compression. The denominator is the square root of the sum of the squares of the local average gray levels of all pixels in the sub-block, with a small positive constant added. This serves to normalize and smooth the data, preventing the numerator from becoming too large due to individual extreme gradient values, while also preventing the denominator from being zero. Therefore, The larger the value, the more abnormal the frequency energy distribution of the sub-block is, and the more severe the compression distortion may be.

[0034] The steps to obtain each parameter are as follows: For sub-blocks Total number of pixels: This parameter is directly determined by the fixed size set when dividing the image into sub-blocks. For example, if the sub-block size is set to 8×8 pixels, then... Each pixel. If the sub-block size is 16×16 pixels, then Each pixel. For a specific application, such as selecting a sub-block size of 16×16, then .

[0035] For sub-blocks The Middle The grayscale value of each pixel: This grayscale value comes from the result of grayscale processing of the subpixel-corrected image data in the previous step. For example, sub-blocks The first in If a pixel has a value of 128 in the grayscale subpixel-corrected image data, then... .

[0036] This parameter represents the grayscale gradient of the pixel in the horizontal direction. The rate of change of grayscale in the horizontal direction. This is typically calculated using gradient operators. For example, the horizontal Sobel operator. For pixels Convolution operations are performed on the 3×3 neighborhood of the pixel. For example, if the pixel The horizontal gradient obtained by applying the Sobel operator to the gray values ​​of the surrounding area is -30. .

[0037] The grayscale gradient of this pixel in the vertical direction: This parameter represents the pixel's grayscale gradient. The rate of change of gray level in the vertical direction. Similar to the horizontal gradient, the vertical Sobel operator can be used. For pixels Convolution operations are performed on the 3×3 neighborhood of the pixel. For example, if the pixel The vertical gradient of the gray values ​​of the surrounding area, calculated using the Sobel operator, is 50. .

[0038] The local average gray level of the pixel's neighborhood: This parameter is the pixel's... The arithmetic mean of the grayscale values ​​of all pixels within a small neighborhood window. For example, for a pixel... Its 3×3 neighborhood (including itself) has gray values ​​of {120, 122, 125, 128, ...}, =130, 132, 135, 138, 140}, then .

[0039] To prevent constants with a denominator of zero: set .

[0040] Calculation process: Taking a 16×16 pixel sub-block as an example, then .set up .

[0041] Only one pixel in the sub-block is calculated Contribution to the numerator and denominator.

[0042] For example, for sub-blocks The first pixel in ): (Grayscale value); (Horizontal Sobel gradient); (Vertical Sobel gradient); (3x3 neighborhood average gray level); Step 1: Calculate the contribution of this pixel to the numerator; Internal logarithmic terms: ; Logarithmic calculation: ; The absolute value of the product with the horizontal gradient: ; This is The result is that the molecule is .

[0043] Step 2: Calculate the denominator for this pixel. Contributions; ; This is The result.

[0044] Step 3: Calculate the denominator; ; Step 4: Calculation (Illustrative) ); ; In practical applications, all sub-blocks need to be included. Sum of the numerator contributions of each pixel, and all 1 pixel Summate, then substitute into the formula. For example, if The sum of the numerators is 15000, and the denominator contains... ,but: ; This result indicates that for the calculated sub-blocks Its frequency energy anomaly value is This value reflects the complexity of the texture within the sub-block and the degree of distortion that may be introduced by factors such as compression. A higher value indicates greater complexity. A value typically indicates strong, irregular gradient changes within the sub-block, which could be a manifestation of compression distortion such as block effect boundaries or ringing. This value will be used to subsequently determine whether the sub-block belongs to the core compression distortion sub-block.

[0045] Based on the frequency energy anomaly values ​​of each marked sub-block calculated in the previous step The system first needs to determine a criterion to filter out the regions with the most severe compression distortion. To do this, the system will count all sub-blocks (or only sub-blocks in the boundary difference marker list, depending on the implementation strategy; here, for example, all sub-blocks are calculated). Frequency energy anomaly (value) and based on these Calculate the arithmetic mean (mean) of all values. ) and standard deviation ( ), determine the threshold Set as "the sum of the mean and standard deviation of the frequency energy anomalies in the entire image sub-block", that is The coefficient "1" here represents a distance of one standard deviation, which can be adjusted according to the sensitivity requirements for distortion detection. For example, if... and ,but Then, the system will iterate through each sub-block. Value, if a certain sub-block's Value greater than this threshold (For example, its) If a sub-block is identified as a compression distortion core sub-block, it is added to the compression distortion core sub-block set. Next, for each sub-block in the compression distortion core sub-block set, the system further analyzes the gradient characteristics of its internal pixels. Specifically, for each pixel within the sub-block, its horizontal grayscale gradient on the sub-pixel-corrected image data (after grayscale conversion) is extracted. and vertical grayscale gradient (These gradient values ​​are calculated) (The gradient data may already be available and can be reused), and calculate the gradient magnitude and direction for each pixel, including the gradient magnitude. gradient direction Then, the pixel gradients are classified according to preset rules based on the principal gradient direction. For example, the gradient direction can be classified. Quantified into several main categories: level-dominant (e.g.) exist or ), vertical dominance (e.g.) exist or ), and others (diagonal or mixed directions), or simple comparisons and The size, if (in A coefficient slightly greater than 1, such as 1.2, is used to enhance the stability of orientation discrimination. If the coefficient is less than 1, the orientation is horizontally dominant, and vice versa. This information, including the position of each compression distortion core sub-block, the gradient magnitude of each pixel within it, and the classification result of the principal gradient direction, together constitute the compression distortion details parameters.

[0046] The steps for obtaining the differentiated weight set are as follows: Based on the compression distortion details parameters, the pixel position coordinates and corresponding pixel gradient magnitude values ​​of all sub-blocks in the compression distortion core sub-block set are extracted. The pixel gradient magnitude values ​​of each sub-block are sorted from largest to smallest. The maximum and average gradient magnitude values ​​after sorting are recorded. Sub-blocks whose maximum pixel gradient magnitude value exceeds twice the average gradient are marked as high distortion sensitive areas, and a list of high distortion sensitive areas is generated. Based on the list of high-distortion sensitive areas, the viewpoint tracking data provided by VR is called to determine the coordinate position of the center area of ​​the field of view where the user's viewpoint is focused. The spatial distance from each sub-block in the high-distortion sensitive area to the center area of ​​the field of view is calculated based on the center area of ​​the field of view. The high-distortion sensitive areas are sorted according to the spatial distance to form a priority sorting list of areas. Based on the regional priority sorting list, a preset processing intensity level is matched for each sub-block in the sorting list to generate a differentiated processing weight set.

[0047] Specifically, based on the compression distortion details parameters obtained in the previous step, which already contain pixel-level gradient information of each sub-block identified as part of the compression distortion core sub-block set, the system first traverses each sub-block in this set. For each sub-block, it extracts the gradient magnitude values ​​of all its internal pixels on the sub-pixel corrected image grayscale from the compression distortion details parameters. These gradient magnitude values ​​were previously obtained by calculating the square root of the sum of the squares of the horizontal and vertical gradients of each pixel. For the currently processed sub-block, the system sorts the collected gradient magnitude values ​​of all pixels within that sub-block in descending order, from the maximum gradient magnitude value to the minimum gradient magnitude value. After sorting, the system records the maximum value of the pixel gradient magnitude within that sub-block, denoted as... And calculate the arithmetic mean of the gradient magnitude values ​​of all pixels within the sub-block, denoted as . This average value is obtained by summing the gradient magnitude values ​​of all pixels within the sub-block and dividing by the total number of pixels in the sub-block (e.g., for a 16×16 pixel sub-block, the total number of pixels is 256). Next, the system will perform a sensitivity assessment on the sub-block, based on the maximum pixel gradient magnitude of that sub-block. Does it exceed the average pixel gradient magnitude within its corresponding sub-block? Twice that, i.e., judging inequalities The coefficient "2" is set empirically to identify sub-blocks with extremely large internal gradient changes, which may contain very abrupt edges or distortions. The basis for setting this coefficient is: by analyzing a large number of image samples containing different degrees of compression distortion, it was observed that when the maximum gradient in a sub-block is much greater than the average gradient (e.g., more than twice), the sub-block has a high probability of being a distortion area that is easily perceived by the human eye. If the above inequality holds, the sub-block is marked as a high distortion sensitive area, and the identifier of the sub-block (e.g., its starting coordinates or index in the image) is added to a new list. After traversing all sub-blocks in the core compression distortion sub-block set, this list containing all the identifiers of sub-blocks marked as high distortion sensitive areas is the high distortion sensitive area label list.

[0048] Based on the list of high-distortion sensitive areas generated in the previous step, the system will next prioritize these areas by combining the user's real-time visual focus information. First, the system calls the real-time viewpoint tracking data provided by the VR headset display device through an interface. This data is usually updated at a certain frequency (e.g., 60 times per second or higher) and provides the two-dimensional coordinates of the user's current gaze point on the display screen. Or its corresponding point in the virtual 3D scene, the system obtains the average gaze point coordinates of the current frame or the recent period, and uses them as the coordinates of the center region of the field of view where the user's view is focused. To simplify the calculation, the coordinates of this center region of the field of view can be directly set as the gaze point. The system either takes the center coordinates of the image sub-block containing the gaze point, or takes the center coordinates of the image sub-block containing the gaze point. Then, the system iterates through each sub-block marked as a high-distortion sensitive region in the high-distortion sensitive region marker list. For each such sub-block, the system determines its representative spatial location in the image, typically the coordinates of the geometric center point of the sub-block. Based on this position and the coordinates of the center region of the field of view that the user's viewpoint is focused on, The spatial distance between the two points is calculated using Euclidean distance, which is obtained by taking the square root of the sum of the squares of the differences between the coordinates of the two points on the horizontal and vertical axes. For example, if the center of the sub-block is... And the center of the field of view is Then the distance After the system calculates a spatial distance value between each sub-block in the high-distortion sensitive area labeling list and the center of its field of view, it sorts the sub-blocks in the list in ascending order based on these calculated spatial distance values. That is, the sub-blocks that are closer to the user's viewpoint focus center are ranked higher and have higher priority. Through this sorting operation, a region priority sorting list is finally formed.

[0049] Based on the region priority sorting list generated in the previous step, which has already sorted all high-distortion sensitive regions in ascending order according to their distance from the user's viewpoint focus center, the system then assigns a preset processing intensity level to each sub-block in this sorting list. These processing intensity levels are a predefined set of discrete or continuous values ​​representing the intensity to be applied in subsequent image processing (such as deblocking filtering). For example, three processing intensity levels can be preset: High (corresponding to a weight value)... ), and (corresponding weight values) ), low (corresponding weight value) These levels and their corresponding weights are set based on extensive visual experiments and user feedback. Considering the human eye's greater sensitivity to image quality in the central region of the field of vision and decreasing sensitivity at the edges, regions closer to the viewpoint center are assigned higher processing intensity, while those farther away are assigned lower. The specific matching rules can be set as follows: Divide the total number of sub-blocks in the region priority ranking list into three equal segments (or according to a specific ratio, e.g., the first 20% as high, the next 30% as medium, and the rest as low). Sub-blocks in the first third of the list (i.e., those closest to the viewpoint center) are assigned a "high" processing intensity level, those in the middle third are assigned a "medium" processing intensity level, and those in the last third (i.e., those farthest from the viewpoint center) are assigned a "low" processing intensity level. As another more refined matching method, a distance threshold can be set, for example, based on the visual angle and screen pixel density under typical VR viewing conditions, the visual angle... and Converted to pixel distance on the screen and If the distance between the sub-block and the center of the field of view (For example If the corresponding pixel is 100, then a "high" processing intensity level is assigned; if (For example If the corresponding pixel count is 300, then assign a "medium" processing intensity level; if If a certain condition is not met, a "low" processing intensity level will be assigned. The system will determine such a processing intensity level and its corresponding numerical weight for each sub-block in the regional priority sorting list. The set of all these sub-block identifiers and their corresponding processing intensities (or weight values) constitutes the differentiated processing weight set.

[0050] The steps for obtaining an anti-compression distortion corrected image are as follows: Based on subpixel corrected image data, the corrected grayscale values ​​of the red, green, and blue channels of all pixels in each sub-block are extracted pixel by pixel. The mean, variance, and peak position of the grayscale distribution of each channel of each sub-block are statistically analyzed to generate a grayscale statistical feature set of the sub-block. Based on the gray-scale statistical feature set of sub-blocks, the differential processing weight set is called to match the regional processing intensity of each sub-block. Based on the regional processing intensity, the filtering radius and filtering intensity of the deblocking filter corresponding to each sub-block are adjusted one by one to form a differential filtering parameter set for each sub-block deblocking filter. Based on the differentiated filtering parameter set, the corresponding filtering parameters are called for each sub-block. The sub-pixel correction grayscale values ​​of each sub-block are filtered by convolution operation. The filtered sub-block image data are then smoothly stitched together to reconstruct all sub-block images and map them back to the original image space coordinates, generating an anti-compression distortion correction image.

[0051] Specifically, based on the subpixel-corrected image data obtained in the aforementioned steps, which includes the corrected grayscale values ​​(e.g., within the range of 0-255) of the red, green, and blue channels of each pixel after subpixel contribution correction, the system first divides the image into multiple sub-blocks according to the same fixed size (e.g., 16×16 pixels) as during compression distortion detection. Then, for each sub-block, the system extracts the corrected grayscale values ​​of the red, green, and blue color channels of all pixels within it, and performs statistical analysis on these extracted corrected grayscale values. Specifically, for the currently processed sub-block, the system calculates its overall average grayscale value, which is obtained by first calculating the brightness value of each pixel within the sub-block, and then calculating the arithmetic mean of the brightness values ​​of all pixels within the sub-block. Simultaneously, the system also calculates the variance of these pixel brightness values ​​within the sub-block to measure the dispersion of grayscale changes within the sub-block. This variance is calculated by calculating the variance of each pixel... The average of the squares of the differences between the brightness value and its average brightness value is obtained. In addition, for the corrected grayscale of each color channel (red, green, blue), the system independently counts its distribution within the sub-block by constructing a grayscale histogram. For example, for an 8-bit grayscale value (0-255), 256 equally wide statistical intervals (bins) can be set. The number of pixels whose corrected grayscale value of each channel falls into each interval within the sub-block is counted. Then, the grayscale value with the most pixels or the median of the grayscale interval is determined from the grayscale histogram of that channel as the peak position of the grayscale distribution of that channel. The above statistical information calculated for each sub-block, including its overall average grayscale, grayscale variance, and the peak positions of its red channel, green channel, and blue channel distributions, is integrated into a structured data record. The collection of these data records for all sub-blocks generates the sub-block grayscale statistical feature set.

[0052] Based on the sub-block grayscale statistical feature set generated in the previous step (although its content is not directly used during the execution of this paragraph, its generation is a prerequisite step), and call the previously constructed differential processing weight set. This differential processing weight set assigns a processing intensity level to each sub-block in the high-distortion sensitive area (for example, high, medium, low, corresponding to numerical weights such as 1.0, 0.7, 0.4 respectively). Next, the system will set differential de-blocking filter parameters for each sub-block that needs to be processed (usually those sub-blocks identified as containing compression distortion). First, the system traverses each sub-block in the differential processing weight set to obtain its corresponding regional processing intensity. This intensity value will directly affect two key parameters of the de-blocking filter: the filtering radius and the filtering intensity. The filtering radius determines the spatial range of the filter's action, that is, centered on the current pixel, how large a neighborhood of pixels the filter will consider; the filtering intensity controls the degree of smoothing. The higher the intensity, the more obvious the smoothing effect, but it may also lead to more loss of image details. The adjustment method is as follows: Set a set of reference filtering parameters, such as a reference filtering radius (for example, 3 pixels) and a reference filtering intensity (for example, a normalized intensity value between 0 and 1, 0.5). These reference values are determined by conducting multiple filtering experiments on typical compressed images and selecting the parameter combinations that can achieve better visual effects under average distortion. Then, according to the regional processing intensity level (high, medium, low) of the current sub-block obtained from the differential processing weight set, apply different adjustment coefficients to adjust these two reference parameters. For example, if the regional processing intensity is "high", the filtering radius is adjusted to (for example , then , which can be rounded to 4 or 5 pixels), and the filtering intensity is adjusted to (for example , then ); if it is "medium", then (for example , pixels), (for example , ); if it is "low", then (for example , , which can be rounded to 2 pixels), (for example , ), and the adjustment coefficient The specific values ​​are obtained by comprehensively optimizing the subjective visual evaluation (such as MOS score) and objective indicators (such as PSNR and SSIM) of the filtering effect at different intensity levels, ensuring that the processing intensity matches the degree of distortion and the characteristics of human eye perception. By setting such a pair of specific filtering radius and filtering intensity values ​​for each sub-block, the system forms a differentiated filtering parameter set for each sub-block deblocking filter.

[0053] Based on the differentiated filtering parameter set of the deblocking filters for each sub-block formed in the previous step, which specifies a specific filtering radius and filtering intensity for each sub-block to be processed, the system then applies the deblocking filters to each sub-block to filter the subpixel-corrected image data. Specifically, for each sub-block, the system first retrieves its corresponding filtering radius and filtering intensity from the differentiated filtering parameter set, and then selects a suitable deblocking filtering algorithm, such as an adaptive mean or Gaussian filter based on convolution operations, or a more advanced conditional filter. The size or influence range of the convolution kernel is adjusted according to the retrieved filtering radius, while the smoothing effect or weight of the filter is set according to the retrieved filtering intensity. The filtering operation is applied to the corrected grayscale values ​​of the red, green, and blue color channels of the subpixel-corrected image data within the sub-block. For example, for each pixel within the sub-block, a weighted average can be performed within its neighborhood defined by the filtering radius, based on the filtering intensity. Both the size of the neighborhood and the filtering process are controlled by these two parameters. After filtering all pixels within a sub-block, since this block processing method may introduce new discontinuities at the boundaries between sub-blocks, it is necessary to perform boundary smoothing stitching on the filtered sub-block image data. This process can be achieved by setting an overlapping region of several pixels at the boundaries of adjacent sub-blocks, and performing a weighted average (alpha mixing) on ​​the corresponding pixel values ​​from the two filtered sub-blocks within the overlapping region. The weights vary with the distance of the pixels from the center of their respective sub-blocks. Alternatively, after all sub-blocks have completed the initial filtering, a slight, directional smoothing filter is applied to the sub-block boundary regions of the entire image to eliminate potential stitching marks. After all sub-blocks have been filtered and their boundaries smoothed, the system reassembles these processed sub-block data according to their spatial positions in the original image and maps them back to the spatial coordinate system of the original image, ultimately generating an anti-compression distortion corrected image that is smooth overall and suppresses compression distortion.

[0054] The steps for obtaining visual salient feature data are as follows: Based on the image with anti-compression distortion correction, the gray values ​​of the red, green and blue channels of each pixel in the image are extracted pixel by pixel. The brightness value and local contrast value corresponding to the gray values ​​of all pixels are calculated respectively. The brightness value and local contrast value are segmented and classified respectively. The number of pixels and the average gray value of each brightness segment and contrast segment are counted to generate statistical features of brightness and contrast segments. Based on the segmented statistical characteristics of brightness and contrast, the relative light intensity contribution factor of each independent light-emitting unit in the sub-pixel spatial contribution distribution map is called to determine the human eye's perception threshold of image information under each brightness and contrast segment. The perceptual threshold is used as a benchmark to determine the adjustment range of pixel grayscale sharpening segment by segment, forming a pixel grayscale sharpening adjustment range mapping relationship. Based on the pixel grayscale sharpening adjustment amplitude mapping relationship, the grayscale value of the corresponding pixel in the anti-compression distortion corrected image is adjusted pixel by pixel, and the local contrast features and edge saliency features of the image pixels after grayscale sharpening adjustment are extracted to obtain visual saliency feature data.

[0055] Specifically, based on the compression-resistant image obtained in the previous step, the system first iterates through every pixel in the image. For each pixel... The system extracts the grayscale values ​​of the red (R), green (G), and blue (B) color channels from the anti-compression distortion corrected image. These grayscale values ​​are typically in the range of 0 to 255. Then, the system calculates the brightness value of the current pixel based on these three channel grayscale values. Simultaneously, the system also calculates the local contrast value of the pixel. This is done by first defining a value based on the current pixel... Take a local neighborhood window centered on the pixel, such as a 7x7 pixel window, and then find the maximum value of the brightness of all pixels within that window. and minimum value Local contrast can be achieved by using If the denominator is zero, the contrast is zero. After calculating the brightness value of all pixels and the contrast value of local areas in the image, the system classifies these brightness values ​​and local area contrast values ​​into segments. For brightness values, for example, they can be divided into 5 fixed segments: [0, 50] (dark area), [51, 100] (secondary dark area), [101, 150] (medium bright area), [151, 200] (secondary bright area), [201, 255] (bright area). For local area contrast values ​​(which are usually between 0 and 1), for example, they can be divided into 4 segments: [0, 0.1] (very low contrast), [0, 0.1] (extremely ... The boundaries and number of these intervals (1, 0.3] (low contrast), (0.3, 0.6] (medium contrast), and (0.6, 1.0] (high contrast) are predetermined based on the target display characteristics and human visual perception characteristics, through analysis of the statistical distribution of a large number of image samples and expert experience. Subsequently, the system counts the number of pixels contained in each brightness segment interval and the average gray value (meaning average brightness value) of these pixels. Similarly, it also counts the number of pixels contained in each contrast segment interval and the average gray value (meaning average brightness value) of these pixels. After summarizing these statistical data, the segmented statistical features of brightness and contrast are generated.

[0056] Based on the segmented statistical features of brightness and contrast generated in the previous step, and by calling the sub-pixel spatial contribution distribution map obtained in an earlier stage (this map records the relative light intensity contribution factors of the three independent light-emitting units of red, green, and blue at each pixel location), , , The system will then determine the perceptual threshold of the human eye for image information under different visual conditions. This perceptual threshold can refer to, for example, the smallest contrast change or the smallest detail size that the human eye can distinguish. The determination process will comprehensively consider the brightness and contrast segments being analyzed, as well as the sub-pixel spatial contribution distribution information related to the pixels within that segment. Specifically, for each combination of brightness and contrast segments (e.g., a mid-brightness and low-contrast segment), the system first obtains a basic perceptual threshold based on an existing visual model. The base threshold primarily depends on the average brightness and contrast of the current segment. Then, the system considers the average sub-pixel spatial contribution factor of the pixels within that segment (e.g., for all pixels within the segment). Calculate the average, and then average the three color averages to obtain a comprehensive contribution factor. This factor is then used to correct the baseline perception threshold, for example, the perception threshold. ,in It is a positive adjustment coefficient (e.g.) ), this coefficient The settings are based on experiments and aim to quantify the effect of subpixel contribution uniformity on the improvement of detail discernibility: higher This indicates that a more ideal subpixel contribution theoretically allows the human eye to perceive more subtle differences in this area, thus lowering the effective perception threshold, and vice versa. This determines the perception threshold for each brightness and contrast segment. Then, the system will use this perception threshold as a benchmark to determine the adjustment range of subsequent pixel grayscale sharpening operations segment by segment. The principle for setting the sharpening amplitude is as follows: for segments with a low perception threshold (i.e., high human eye sensitivity), the sharpening amplitude should be appropriately reduced to avoid over-sharpening artifacts; for segments with a high perception threshold (low human eye sensitivity), the sharpening amplitude can be appropriately increased to enhance details. For example, the adjustment amplitude can be set to be proportional to the reciprocal of the perception threshold, or it can be set using a preset non-linear mapping function. get For example, if If it is lower (e.g., 0.02), then It is also lower (e.g., 0.3); if If it is higher (e.g., 0.1), then The values ​​are also relatively high (e.g., 1.2). These specific amplitude values ​​and their mapping relationship with the perceptual threshold were determined by conducting multiple sharpening experiments on representative images, and having observers rate the visual effects under different parameter combinations (e.g., sharpness, naturalness, presence of noise or halo, etc.). The optimal parameter configuration was then selected and fixed. Finally, each combination (brightness segment, contrast segment) was associated with its corresponding sharpening adjustment amplitude. When these are linked together, a mapping relationship is formed for the adjustment range of pixel grayscale sharpening.

[0057] Based on the pixel grayscale sharpening adjustment range mapping relationship established in the previous step, which pre-sets the corresponding sharpening adjustment range for different brightness and contrast segment combinations, the system will then perform grayscale value sharpening adjustment on each pixel in the anti-compression distortion rectified image one by one. Specifically, for any pixel in the anti-compression distortion rectified image... First, calculate its current brightness value. and local area contrast value Based on these two values, the brightness segment and contrast segment to which the pixel belongs are determined. Then, the sharpening adjustment range of the corresponding segment combination is obtained by querying the pixel grayscale sharpening adjustment range mapping relationship. After obtaining the adjustment range, an unsharpened mask is used to adjust the grayscale values ​​of the red, green, and blue channels of the pixel. The basic process of the unsharpened mask is to first smooth the grayscale values ​​of each channel of the original pixel (e.g., Gaussian blur) to obtain the grayscale values ​​of the corresponding channels in the blurred image. Then, calculate the difference (high-frequency component) between the original grayscale value and the blurred grayscale value, and multiply this difference by the retrieved sharpening adjustment range. Finally, it is added back to the original grayscale value, that is... To ensure that the adjusted grayscale values ​​remain within a valid range (e.g., 0-255), after performing this sharpening adjustment on all pixels in the image, the system will further extract the updated local contrast features and newly added edge saliency features from these grayscale-sharpened image pixels. The method for extracting local contrast features is the same as that for extracting them before sharpening, but this time it is performed on the sharpened image data. Edge saliency features are obtained by calculating the gradient magnitude of each pixel in the sharpened image (e.g., using the Sobel operator to calculate the horizontal and vertical gradients of the sharpened brightness image and then obtaining its vector magnitude) or by using other edge detection operators (e.g., the edge intensity map output by the Canny edge detector). By integrating these sharpened local contrast values ​​and edge saliency values ​​(e.g., gradient magnitude) extracted from each pixel (or its region), the visual saliency feature data is obtained.

[0058] The steps for acquiring visually optimized images are as follows: Based on visual saliency feature data, local contrast features and edge saliency features of each pixel are extracted pixel by pixel. The feature intensity value of each pixel and the corresponding position coordinates of the pixel in the image coordinate space are recorded. The high, medium and low saliency regions of the pixels are marked according to the feature intensity value, and a visual saliency marking map is generated. Based on the visual saliency marker map, the gray values ​​of the red, green, and blue channels of the corresponding positions in the anti-compression distortion corrected image are called. Texture enhancement is performed on the gray values ​​of pixels in high saliency areas, edge enhancement is performed on the gray values ​​of pixels in medium saliency areas, and noise suppression is performed on the gray values ​​of pixels in low saliency areas, thus forming regional gray-scale adjusted image data. Based on the image data with grayscale adjustment in different regions, the high, medium and low saliency regions after grayscale adjustment are smoothly and gradually merged pixel by pixel. The grayscale boundaries of each channel of the merged image are then smoothly interpolated to reconstruct continuous and uniform color image data, resulting in a visual recognition optimized image.

[0059] Specifically, based on the visual salient feature data obtained in the previous step, which provides local contrast feature values ​​and edge saliency feature values ​​for each pixel in the image, the system first integrates these two features pixel by pixel to obtain a comprehensive feature intensity value. For example, a weighted average method can be used to calculate it, and the formula is: Comprehensive Feature Intensity Value ,in It is a pixel Local contrast feature values, It is its marginal significance eigenvalue (e.g., normalized gradient magnitude), weight and (For example, setting them to 0.6 and 0.4 respectively) was determined based on the perceived importance of different visual elements to the human eye. This was achieved through a series of subjective evaluation experiments using images containing different scene content (e.g., asking observers to mark the most attention-grabbing areas in the images, then analyzing the correlation between the contrast and edge characteristics of these areas and the ratings). This ensures... The system records the combined feature intensity value of each pixel and its position coordinates in the image. Next, based on the combined feature intensity values ​​of all pixels, the pixels are divided into three saliency levels: high, medium, and low. This division is achieved by setting two thresholds. and To achieve this, the two thresholds are determined as follows: First, calculate the complete statistical histogram of the comprehensive feature intensity values ​​of all pixels in the image. Then, based on the cumulative distribution of this histogram, arrange the comprehensive feature intensity values ​​from high to low and set... To ensure that the top 20% of pixels with the highest comprehensive feature intensity values ​​in the image (this percentage can be adjusted according to the sparsity requirements of the highly saliency region in the application scenario, for example, within a range of 10%-30%) are classified as "highly saliency" threshold points, similarly, a threshold is set... To ensure that the 30% of pixels with the lowest overall feature intensity values ​​(e.g., an adjustment range of 20%-40%) are classified as "low significance" threshold points, the remaining pixels (i.e., those with overall feature intensity values ​​between...) are... and Values ​​between 0 and 1 are marked as "medium significance". For example, if the combined feature intensity values ​​of all pixels (e.g., normalized to the 0-1 range), when calculated and sorted in descending order, have a 20th percentile value of 0.75 and a 70th percentile value (i.e., the value corresponding to the 30th percentile counting from the bottom up) of 0.30, then... , Pixels with a comprehensive feature intensity value greater than 0.75 are marked as highly significant, those between 0.30 and 0.75 are marked as moderately significant, and those less than or equal to 0.30 are marked as lowly significant. By assigning such a salience level (high, medium, low) to each pixel in the image, a visual salience label map is generated.

[0060] Based on the visual saliency map generated in the previous step, which labels each pixel in the image with its saliency level (high, medium, low), the system calls upon the compression-resistant correction image obtained in an earlier stage (i.e., the image version that has undergone deblocking filtering but has not yet undergone visual saliency sharpening). For regions with different saliency levels, the system performs differentiated image enhancement or adjustment operations on the grayscale values ​​of the red, green, and blue channels of the corresponding pixels in the compression-resistant correction image. Specifically: For pixels marked as "high saliency regions" in the visual saliency map, the system performs texture enhancement processing on their original red, green, and blue channel grayscale values. For example, an adaptive contrast enhancement algorithm based on local standard deviation can be used. For each pixel, the grayscale standard deviation is calculated within its neighborhood (e.g., 5×5 pixels). If the standard deviation is large (indicating rich texture), the contrast of that neighborhood is further stretched, and the enhancement coefficient is increased. (For example, set between 1.2 and 1.5, with the specific value selected through testing, such as 1.3) Control the intensity of the enhancement. For pixels marked as "medium saliency areas," the system performs edge enhancement processing on their grayscale values. For example, an enhanced unsharpened mask technique can be applied, with a small sharpening radius (e.g., 1-2 pixels) and a sharpening amount... The value should be moderate (e.g., between 0.8 and 1.2, such as 1.0), and can be combined with edge direction information to enhance only the main edge direction to avoid blurring non-edge areas. For pixels marked as "low saliency areas," the system performs noise suppression processing on their grayscale values, for example, by applying a median filter (window size such as 3×3 pixels) or a bilateral filter (spatial domain standard deviation). For example, 2, range standard deviation For example, 20), to smooth noise and preserve major edges, the specific algorithm parameters selected for these operations (texture enhancement, edge strengthening, noise suppression) are as described above. , The filter window size and standard deviation are all best practice values ​​that have been optimized and determined in advance by processing a large number of sample images and evaluating their visual effects (including sharpness, naturalness, and distortion suppression). This ensures that the processing method for each type of region matches its visual importance and content characteristics. After all pixels undergo specific grayscale adjustments in their corresponding salient regions, the adjusted red, green, and blue channel data together form the regional grayscale adjusted image data.

[0061] Based on the segmented grayscale adjusted image data generated in the previous step, where different saliency regions (high, medium, and low) have already undergone specific grayscale adjustments (texture enhancement, edge strengthening, or noise suppression), to avoid visually abrupt transitions or artifacts at the boundaries of these different processed regions, the system next needs to perform a smooth gradient fusion process pixel by pixel on these grayscale-adjusted high, medium, and low saliency regions. Specifically, the boundary bands between regions of different saliency levels can be identified first. For example, the range extending outward from the high saliency region by 2 to 3 pixels can be defined as the transition band with the medium saliency region, and a similar transition band can be defined between the medium and low saliency regions. For pixels within these transition bands, the final red, green, and blue channel grayscale values ​​will be obtained by weighted averaging between the processing results in the high saliency region and the medium saliency region (or the medium and low saliency regions). The change is smooth, for example, for the region of high significance. Transition to the mesosignificance region A pixel in the transition band, its final value Weight The smooth transition from 1 near the highly saliency region to 0 near the moderately saliency region can be achieved using linear interpolation or a weighted distribution based on a Gaussian function shape. Furthermore, to ensure a natural and smooth grayscale boundary across all color channels, the system performs additional smoothing interpolation on any subtle grayscale jumps or boundaries in the fused image caused by different previous operation intensities. This may involve applying a very slight, directional smoothing filter (e.g., a Gaussian blur with a kernel size of 3x3 and a standard deviation of 0) to the boundaries of the identified saliency regions and their immediate vicinity. If set to a small value, such as 0.5, the filter will smooth along the boundary direction rather than perpendicular to the boundary, in order to maintain the sharpness of the edge while softening the transition. After the above smooth gradient fusion and boundary smoothing interpolation processing, the system recombines the grayscale data of the red, green and blue channels of all pixels and reconstructs a visually continuous and uniform color image data, which is the final visual recognition optimized image.

[0062] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A virtual reality (VR) image recognition method based on Micro-LED, characterized in that, Includes the following steps: Based on the input image of the VR display, the light intensity contribution and spatial arrangement of the sub-pixels of the Micro-LED display, namely the independent light-emitting units of red, green and blue, are analyzed to generate a sub-pixel spatial contribution distribution map. The pixel data is recalculated by combining the original image data with the contribution value of each independent light-emitting unit in the sub-pixel spatial contribution distribution map to obtain sub-pixel corrected image data. Based on the subpixel-corrected image data, the presence of block boundary features is detected, anomalies in frequency energy distribution are analyzed, the severity of compression distortion is determined, compression distortion detail parameters are obtained, and based on the compression distortion detail parameters, differential processing coefficients are set for different regions of the image to obtain a differential processing weight set. Based on the subpixel-corrected image data and the differential processing weight set, the parameters of the deblocking filter are adjusted by calling the region processing coefficients in the differential processing weight set and applied to generate an anti-compression distortion corrected image. Based on the anti-compression distortion corrected image and the subpixel spatial contribution distribution map, the human eye's perception threshold for information under different brightness and contrast values ​​in the anti-compression distortion corrected image is calculated. Based on the perception threshold and referring to the weight of each independent luminous unit in the subpixel spatial contribution distribution map, sharpening adjustment is performed to obtain visual salient feature data. The visual salient feature data is then fused with the anti-compression distortion corrected image to generate a visual recognition optimized image.

2. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the sub-pixel spatial contribution distribution map are as follows: Based on the input image of the VR display, the coordinate position of each pixel is located in the image rendering matrix. The geometric centroid coordinates, luminous area and brightness value of the corresponding red light-emitting unit on the Micro-LED display are matched pixel by pixel. The Euclidean distance from the centroid of each light-emitting unit to the center of the target pixel is measured to form a set of spatial brightness parameters of the red light-emitting unit. Based on the set of spatial brightness parameters of the red light-emitting unit, calculate the relative light intensity contribution factor of the red light-emitting unit to the pixel; Based on the relative light intensity contribution factor, the green and blue light-emitting units are also calculated using the relative light intensity contribution factor formula. Channel fusion and pixel coordinate reverse mapping are performed on the three-color contribution factor of each pixel to generate a sub-pixel spatial contribution distribution map.

3. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for acquiring the subpixel-corrected image data are as follows: Based on the original image data, each pixel of the original image data is scanned pixel by pixel, the initial gray values ​​of the red, green and blue channels corresponding to the pixel are extracted, and the initial gray value matrix of the three channels of each pixel is established to obtain the original pixel gray value matrix. Based on the original pixel grayscale matrix, according to the sub-pixel spatial contribution distribution map, and referring to the relative light intensity contribution factors of the three independent light-emitting units of red, green and blue at each pixel position, the product of the original grayscale value and the corresponding channel relative light intensity contribution factor is calculated for each channel, and the product results of the three channels are normalized to form the sub-pixel channel grayscale correction matrix. Based on the subpixel channel grayscale correction matrix, the corrected grayscale values ​​of the red, green, and blue channels are recombined pixel by pixel, mapped back to the image coordinate system point by point, the image three-channel grayscale correction data is reconstructed, and the image color channels are overlapped and fused to obtain subpixel corrected image data.

4. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the compression distortion details parameters are as follows: Based on the subpixel-corrected image data, the image is divided into multiple sub-blocks of fixed size. The boundary pixel column between each sub-block and its adjacent sub-blocks on the right and below is traversed. The absolute value sequence of the difference between the gray values ​​of corresponding pixels on both sides of the boundary line is calculated, and the absolute value sequence is averaged by a sliding window. If the average gray value difference in three consecutive windows is greater than twice the average gray value difference of the entire image, the sub-block is marked as a sub-block with significant boundary differences, and a boundary difference label list is generated. Based on the boundary difference marker list, calculate the frequency energy anomaly value for each marked sub-block; Based on the frequency energy anomalies, all sub-blocks whose frequency energy anomalies are greater than the sum of the mean and standard deviation of the frequency energy anomalies of the entire sub-block are set as the compression distortion core sub-block set. The horizontal and vertical grayscale gradient amplitudes of all pixels in each sub-block of the compression distortion core sub-block set are extracted, and the pixel gradients are classified according to the principal gradient direction to obtain the compression distortion detailed parameters.

5. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the differentiated processing weight set are as follows: Based on the compression distortion details parameters, the pixel position coordinates and corresponding pixel gradient magnitude values ​​of all sub-blocks in the compression distortion core sub-block set are extracted. The pixel gradient magnitude values ​​of each sub-block are sorted from largest to smallest. The maximum and average gradient magnitude values ​​after sorting are recorded. Sub-blocks whose maximum pixel gradient magnitude value exceeds twice the average gradient are marked as high distortion sensitive regions, and a high distortion sensitive region marker list is generated. Based on the high-distortion sensitive area marker list, the viewpoint tracking data provided by VR is called to determine the coordinate position of the center area of ​​the field of view where the user's viewpoint is focused. The spatial distance from each sub-block in the high-distortion sensitive area to the center area of ​​the field of view is calculated based on the center area of ​​the field of view. The high-distortion sensitive areas are sorted according to the spatial distance to form a region priority sorting list. Based on the region priority sorting list, a preset processing intensity level is matched for each sub-block in the sorting list to generate a differentiated processing weight set.

6. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the compression-resistant image are as follows: Based on the subpixel corrected image data, the corrected grayscale values ​​of the red, green, and blue channels of all pixels in each sub-block are extracted pixel by pixel. The mean, variance, and peak position of the grayscale distribution of each channel of each sub-block are statistically analyzed to generate a grayscale statistical feature set of the sub-block. Based on the gray-scale statistical feature set of the sub-blocks, the differential processing weight set is called to match the regional processing intensity of each sub-block, and the filtering radius and filtering intensity of the deblocking filter corresponding to each sub-block are adjusted one by one based on the regional processing intensity to form a differential filtering parameter set for each sub-block deblocking filter. Based on the differentiated filtering parameter set, the corresponding filtering parameters are called for each sub-block, and the sub-pixel correction grayscale values ​​of each sub-block are filtered by convolution operation. The filtered sub-block image data are then smoothly stitched together to reconstruct all sub-block images and map them back to the original image space coordinates, generating an anti-compression distortion correction image.

7. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the visual salient feature data are as follows: Based on the anti-compression distortion corrected image, the gray values ​​of the red, green and blue channels of each pixel in the image are extracted pixel by pixel. The brightness value and local area contrast value corresponding to the gray values ​​of all pixels are calculated respectively. The brightness value and local area contrast value are segmented and classified respectively. The number of pixels and the average gray value of each brightness segment and contrast segment are counted to generate the brightness and contrast segment statistical features. Based on the segmented statistical characteristics of brightness and contrast, the relative light intensity contribution factor of each independent light-emitting unit in the sub-pixel spatial contribution distribution map is called to determine the human eye's perception threshold of image information under each brightness and contrast segment. The perceptual threshold is used as a benchmark to determine the adjustment range of pixel grayscale sharpening segment by segment, forming a pixel grayscale sharpening adjustment range mapping relationship. Based on the pixel grayscale sharpening adjustment amplitude mapping relationship, the grayscale value of the corresponding pixel in the anti-compression distortion corrected image is adjusted pixel by pixel, and the local contrast features and edge saliency features of the image pixels after grayscale sharpening adjustment are extracted to obtain visual saliency feature data.

8. The virtual reality (VR) image recognition method based on Micro-LED according to claim 1, characterized in that, The steps for obtaining the visual recognition optimized image are as follows: Based on the aforementioned visual saliency feature data, local contrast features and edge saliency features of each pixel are extracted pixel by pixel. The feature intensity value of each pixel and the position coordinates of the corresponding pixel in the image coordinate space are recorded. The high, medium and low saliency regions of the pixels are marked according to the magnitude of the feature intensity value to generate a visual saliency marking map. Based on the visual saliency marker map, the gray values ​​of the red, green, and blue channels of the corresponding positions in the anti-compression distortion corrected image are called. Texture enhancement is performed on the gray values ​​of pixels in high saliency areas, edge enhancement is performed on the gray values ​​of pixels in medium saliency areas, and noise suppression is performed on the gray values ​​of pixels in low saliency areas, thus forming regional gray-scale adjustment image data. Based on the grayscale adjusted image data, the high, medium, and low saliency regions after grayscale adjustment are smoothly and gradually merged pixel by pixel, and the grayscale boundaries of each channel of the merged image are smoothly interpolated to reconstruct continuous and unified color image data, thus obtaining a visual recognition optimized image.

9. A virtual reality (VR) image recognition system based on the Micro-LED-based VR image recognition method according to any one of claims 1-8, characterized in that, include: The sub-pixel analysis module analyzes the light intensity contribution and spatial arrangement of the sub-pixels of the Micro-LED display screen, i.e., the light-emitting units of red, green and blue independent light-emitting units, based on the input image of the VR display. It generates a sub-pixel spatial contribution distribution map, and recalculates the pixel data by combining the original image data with the contribution value of each independent light-emitting unit in the sub-pixel spatial contribution distribution map to obtain sub-pixel corrected image data. The compression distortion analysis module, based on the subpixel corrected image data, detects the presence of block boundary features, analyzes anomalies in frequency energy distribution, determines the severity of compression distortion, obtains compression distortion detail parameters, and sets differential processing coefficients for different regions of the image based on the compression distortion detail parameters to obtain a differential processing weight set; The deblocking filter module, based on the subpixel corrected image data and the differential processing weight set, calls the region processing coefficients in the differential processing weight set to adjust the parameters of the deblocking filter and applies them to generate an anti-compression distortion corrected image. The visual perception enhancement module calculates the human eye's perception threshold for information at different brightness and contrast values ​​in the anti-compression distortion corrected image based on the anti-compression distortion corrected image and the sub-pixel spatial contribution distribution map. Based on the perception threshold and referring to the weight of each independent luminous unit in the sub-pixel spatial contribution distribution map, it performs sharpening adjustment to obtain visual salient feature data. The visual salient feature data is then fused with the anti-compression distortion corrected image to generate a visual recognition optimization image.

Images