Image processing method and display device

Abstract

The application discloses an image processing method and a display device. The display device has a lenticular lens array. The method comprises: performing edge detection processing on a to-be-displayed image to obtain edge information; determining an edge band and a soft mask of the edge band according to the edge information; determining filter kernel parameters of an anisotropic filter kernel according to a scale feature of a closed region corresponding to the edge information and a structure parameter of the lenticular lens array; performing filter processing on the to-be-displayed image in the edge band according to the filter kernel parameters to obtain a filtered image; and performing pixel-by-pixel fusion processing on the filtered image and the to-be-displayed image in the edge band according to the soft mask to obtain a target image for display. The application can adaptively and selectively suppress the local high-frequency structure most related to jaggies without significantly sacrificing the global definition, and can also take into account the directionality and scale dependence of the lenticular display, thereby effectively improving the display effect.

Description

Technical Field

[0001] This application relates to the field of display technology, specifically to an image processing method and a display device. Background Technology

[0002] Cylindrical lens 3D displays achieve a naked-eye stereoscopic effect by superimposing a cylindrical lens array in front of the display panel, multiplexing adjacent pixel columns or sub-pixels according to a preset parallax direction to different viewing angles. While this structure brings multi-view spatial multiplexing, it also introduces an additional spatial sampling process compared to traditional 2D display links: the lens pitch and tilt angle optically redistribute pixels on the screen, causing the overall modulation transfer function (MTF) of the display device to exhibit significant directionality and periodicity. When displaying general 2D content (especially high-contrast text or graphic edges), the source image is usually not pre-filtered and resampled for the lens coordinate domain. Superimposed scaling and sharpening processing in the display link can easily produce spatial aliasing and moiré interference, resulting in jagged edges, color fringing, and flickering, affecting the display effect. Summary of the Invention

[0003] This application provides an image processing method and a display device that adaptively and selectively suppresses the local high-frequency structures most relevant to aliasing, while taking into account the directionality and scale dependence of cylindrical lens displays, effectively improving the display effect.

[0004] In a first aspect, embodiments of this application provide an image processing method applied to a display device having a cylindrical lens array. The method includes: Edge detection processing is performed on the image to be displayed to obtain edge information; Based on the edge information, an edge band and a soft mask of the edge band are determined; wherein, the soft mask is used to indicate the fusion weight of each pixel within the edge band; Based on the scale characteristics of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array, the filter kernel parameters of the anisotropic filter kernel are determined; wherein, the smoothing intensity of the anisotropic filter kernel in the fringe direction perpendicular to the cylindrical lens array is greater than the smoothing intensity of the anisotropic filter kernel in the fringe direction parallel to the cylindrical lens array. According to the filtering kernel parameters, the image to be displayed is filtered within the edge band to obtain a filtered image; According to the soft mask, the filtered image and the image to be displayed are fused pixel by pixel within the edge band to obtain the target image for display.

[0005] Secondly, embodiments of this application provide a display device having a cylindrical lens array, and the display device includes an image processing module for performing the image processing method described above.

[0006] In summary, the technical solution provided in this application involves edge detection processing of the image to be displayed to obtain edge information; determining the edge band and its soft mask based on the edge information; determining the filtering kernel parameters of the anisotropic filtering kernel based on the scale features of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array; filtering the image to be displayed within the edge band according to the filtering kernel parameters to obtain a filtered image; and performing pixel-by-pixel fusion processing on the filtered image and the image to be displayed within the edge band according to the soft mask to obtain the target image for display. Based on this, addressing the issues of edge jaggedness artifacts, color fringing, and flickering generated when displaying two-dimensional content using cylindrical lens 3D displays, this application embodiment, by processing only within the edge band, combining local image scale features and cylindrical lens structural parameters for anisotropic smoothing filtering, and achieving pixel-by-pixel fusion within the edge band using a soft mask, can suppress jaggedness while preserving the main image outline, stripe direction texture, and subjective sharpness as much as possible. This achieves adaptive and selective suppression of the local high-frequency structures most relevant to jaggedness without significantly sacrificing global sharpness, while also considering the directionality and scale dependence of cylindrical lens displays, effectively improving the display effect. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram illustrating the mechanism of edge serrations provided in an embodiment of this application; Figure 2 This is a comparative schematic diagram of a display effect provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a display device provided in an embodiment of this application; Figure 4 This is a schematic flowchart of an image processing method provided in an embodiment of this application; Figure 5 This is a flowchart illustrating another image processing method provided in an embodiment of this application; Figure 6 This is a schematic diagram illustrating the effect of image processing provided in an embodiment of this application; Figure 7This is a schematic diagram illustrating the effect of another image processing method provided in an embodiment of this application. Detailed Implementation

[0009] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. The described technical solutions are for illustrative purposes only and should not be construed as limiting the scope of protection of this application.

[0010] Furthermore, in the embodiments of this application, "multiple" refers to two or more. The terms "first" and "second," etc., in the embodiments of this application are used to distinguish different technical features and do not indicate any order, quantity, or importance.

[0011] The various embodiments provided in this application are similar, and features in different embodiments can be combined with each other.

[0012] The order in which the following embodiments are described is not intended to limit the preferred order of the embodiments.

[0013] As a basic embodiment of this application, in order to solve the technical problems in the background art, the following types of solutions exist: (1) Global low-pass / blur preprocessing: Gaussian or box low-pass is applied to the entire image before it is sent to the screen. Although this solution can reduce high-frequency aliasing to a certain extent, it will generally damage the sharpness, making the strokes of the text appear blurry and significantly reducing the user's subjective sharpness; (2) Edge-selective blurring with fixed kernel: The edge is detected first and then blurred or mixed with a fixed kernel size. This scheme does not consider the edge geometry, local structure (closed / open), directionality and spatial variation of lens parameters, and often results in over-smearing or under-suppression. (3) Subpixel / font rendering and screen parameter coarse coordination: such as rendering based on RGB subpixels or font fine-tuning (hinting). The subpixel direction and sampling kernel assumed by this scheme are usually inconsistent with the lens tilt angle / pitch, and have poor adaptability to individual deviations and viewing distance changes of different devices; (4) Equipment-side calibration and view rearrangement: Lens parameters are calibrated at the factory and rearrangement / pre-filtering is performed at the hardware scale. This approach is usually effective for view interpolation of 3D content, but it still lacks targeted, fine-grained edge anti-aliasing control for 2D UI (User Interface) / text from any source.

[0014] Therefore, how to adaptively and selectively suppress the local high-frequency structures most relevant to aliasing without significantly sacrificing overall sharpness, while taking into account the directionality and scale dependence of cylindrical lens displays, remains a technical challenge for the industry.

[0015] In view of this, embodiments of this application provide an image processing method and a display device, which provide an algorithm to improve edge aliasing and jagged edges when displaying two-dimensional content or mixed two-dimensional / three-dimensional switching content on a cylindrical lens naked-eye 3D display without modifying any hardware. This algorithm can be directly integrated into the display device without additional computing resources and processing units.

[0016] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating the mechanism of edge serrations provided in an embodiment of this application. Figure 2 This is a comparative schematic diagram of a display effect provided in an embodiment of this application. For example... Figure 1 As shown, the gray area on the left is the dark area, and the white area on the right is the bright area. A vertical ideal edge should be formed between the two. The cylindrical lens array covering the screen is arranged obliquely at an angle θ. Its optical sampling direction is inconsistent with the direction of the image edge, which causes the pixels to be incorrectly assigned to different viewing angles, forming a jagged area near the edge. Figure 2 A comparison was made between the original image and the image displayed after passing through the cylindrical lens array: the original image showed ideal vertical black rectangular bars, while the image displayed after passing through the cylindrical lens array exhibited a jagged distortion with diagonal stripes. For example... Figure 1 As shown, in this embodiment, an elliptical Gaussian kernel is placed in the jagged area, and the major axis direction is set to be perpendicular to the stripe direction of the cylindrical lens array to apply stronger smoothing, while the minor axis direction is set to be parallel to the stripe direction of the cylindrical lens array to retain contour details, thereby suppressing jaggedness while maintaining image sharpness as much as possible.

[0017] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of a display device provided in an embodiment of this application. Figure 3 As shown, the display device 300 has a cylindrical lens array 010, and the display device 300 may include an image processing module 310.

[0018] The display device 300 in this embodiment has a cylindrical lens array 010, which enables naked-eye 3D display effect and supports high-quality display of 2D content.

[0019] The display device 300 in this embodiment has an image processing module 310 upstream of the display link. This module performs source-end edge anti-aliasing processing on the image to be displayed, thereby eliminating or suppressing edge aliasing artifacts introduced by the cylindrical lens array 010 before the image data enters the subsequent display link. This source-end processing method requires no hardware modification and can be directly integrated into existing display devices, offering advantages such as low overhead and easy deployment. The image processing module 310 can execute the image processing method provided in any embodiment of this application, as described below. Furthermore, the image processing module 310 can support the processing of various display content, including purely two-dimensional content, purely three-dimensional content, and content that switches between two-dimensional and three-dimensional modes.

[0020] It should be understood that the image processing module 310 is functionally an independent image processing unit, and its physical implementation can be flexibly deployed, which is not limited in this application embodiment. In some embodiments, the image processing module 310 can be integrated inside the display module of the display device 300, directly processing the input image data as a built-in image processing unit without relying on external computing resources; or, the image processing module 310 can be deployed in the GPU (Graphics Processing Unit) or display controller of the display device 300 through a computing shader, utilizing the parallel computing capabilities of the GPU to efficiently process the entire frame image. This deployment method can make full use of the computing resources of existing display control hardware, avoid adding additional processing chips, and help reduce system cost and power consumption.

[0021] During operation, the image processing module 310 can acquire relevant prior information about the panel to support the parameter configuration of the anti-aliasing algorithm. This prior information may include, but is not limited to, pixel format, resolution, pitch and tilt angle of the cylindrical lens array 010, etc. Among them, pitch and tilt angle are key structural parameters of the cylindrical lens array 010, which directly affect the orientation and scale settings of the filter kernel. Pitch determines the period width of the cylindrical lens unit and is positively correlated with the edge band width and filtering intensity; tilt angle determines the extension direction of the cylindrical lens stripes and is directly related to the rotation direction of the anisotropic filter kernel. This prior information may come from the factory calibration during the production stage of the display device 300 and be stored in the display device 300 in the form of firmware parameters or configuration files for the image processing module 310 to access.

[0022] In some embodiments, such as Figure 3 As shown, the display device 300 may further include: an image providing module 320, a display processing module 330, and a display panel 340.

[0023] The image providing module 320 is connected to the image processing module 310 and serves as the image input terminal of the display device 300. Located upstream of the image processing module 310, it can provide the image to be displayed to the image processing module 310. At the system level, the image providing module 320 can correspond to the image compositing unit at the application layer and operating system layer. In modern display systems, applications and operating systems render multiple windows, layers, or rendering surfaces (including but not limited to UI elements, text content, video frames, 3D scenes, etc.) separately, and then composite them into a complete frame image with a target resolution through a window manager or compositor. This composited complete frame image is the image to be displayed. Based on this, in some embodiments, the image providing module 320 can also be used to: composite multiple windows or layers into a complete frame image with a target resolution to obtain the image to be displayed.

[0024] The display processing module 330 is connected to the image processing module 310 and is located downstream of the image processing module 310. It can be used to process the target image output by the image processing module 310 to obtain panel driving signals. The display processing module 330 acts as a bridge connecting the source image processing and the terminal physical display, undertaking key functions such as pixel data conversion, transmission format encapsulation, and panel timing control. In some embodiments, such as... Figure 3 As shown, the display processing module 330 may include: a display control submodule 331, a color processing submodule 332, a link coding submodule 333, and a timing control submodule 334.

[0025] The display control submodule 331 is the entry unit of the display processing module 330, connected to the image processing module 310. It can be used to execute the image kernel operations of the image processing module 310 through the computation shader and output pixel data based on the target image. The display control submodule 331 can be implemented by a GPU or other display processing unit (DPU). GPUs have powerful parallel computing capabilities, and their computation shaders can perform operations on multiple pixels simultaneously, making them suitable for various image kernel operations, such as, but not limited to, gradient calculation, morphological operations, Gaussian kernel convolution, and pixel fusion. By mapping the algorithm logic of the image processing module 310 to a computation shader program, anti-aliasing processing can be efficiently completed using existing hardware resources without adding additional hardware.

[0026] The color processing submodule 332 is connected to the display control submodule 331 and can be used to convert pixel data to the native color gamut and electro-optical characteristics of the display panel 340 to obtain color-corrected pixel data. The display panel 340 has specific color space characteristics (such as native color gamut, gamma response curve, electro-optical conversion function, etc.), which often differ from the standard color space (such as sRGB) of the source image. If the source pixel data is directly sent to the display panel 340 without color adaptation, it will lead to color distortion, brightness deviation, and decreased visual quality. Therefore, in this embodiment, the color processing submodule 332 performs color gamut mapping, gamma correction, tone mapping, and dithering to accurately convert the pixel data to the color space of the display panel 340 to obtain color-corrected pixel data.

[0027] The link encoding submodule 333 is connected to the color processing submodule 332 and can be used to package the color-corrected pixel data into the timing and transmission format of the display link, resulting in an encoded pixel stream. Modern display systems typically use high-speed serial interfaces to transmit pixel data, which require data to be encapsulated according to specific timing and protocol formats. The link encoding submodule 333 is responsible for organizing the pixel data into transmission units that conform to the interface specifications, including, for example, inserting pixel clocks, synchronization signals (horizontal synchronization / vertical synchronization), data packet headers, error correction codes, encryption verification, and other control information. In addition, the link encoding submodule 333 can also manage the bandwidth allocation and transmission timing of the display link, ensuring that the pixel data is transmitted to the downstream timing control submodule 334 at the correct rate and order.

[0028] The timing control submodule 334 is connected to the link encoding submodule 333 and the display panel 340, located downstream of the link encoding submodule 333 and upstream of the display panel 340. It is used to unpack the pixel stream into in-panel timing and generate panel drive signals based on this timing. The timing control submodule 334 can be implemented as a timing controller (TCON) chip, serving as the direct drive front-end of the display panel 340. The timing control submodule 334 decodes and verifies the received pixel stream, extracts pixel data and synchronization information, and then generates corresponding timing control signals according to the drive protocol of the display panel 340. The in-panel timing can include source drive timing (controlling the voltage output of the data lines) and gate drive timing (controlling the row-by-row selection of the scan lines). The timing control submodule 334 sends the voltage data of each pixel to the corresponding source drive circuit according to a precise timing relationship and selects the pixel array row by row through the gate scan signal, achieving row-by-row refresh of the entire frame image. The panel drive signals generated by the timing control submodule 334 may include analog voltage signals and digital control signals. These signals are transmitted to the drive integrated circuit of the display panel 340 to control the voltage of each pixel electrode, thereby modulating the light output intensity of the pixel.

[0029] Display panel 340 is the terminal output unit of display device 300, connected to display processing module 330 (such as timing control submodule 334 in display processing module 330), and display panel 340 includes lenticular lens array 010. Display panel 340 can be used to convert panel driving signals into light output, and distribute pixel emission directions to different viewing angles through lenticular lens array 010.

[0030] The display panel 340 in this embodiment can be implemented using various display technologies, such as, but not limited to, liquid crystal display (LCD) panels, organic light-emitting diode (OLED) panels, and quantum dot organic light-emitting diode (QD-OLED) panels. Different types of display panels have different light-emitting principles and driving methods, but all of them modulate light output by controlling the voltage of the pixel electrodes.

[0031] The cylindrical lens array 010 is a key optical component of the display panel 340, typically bonded or integrated into the light-emitting side of the display panel 340 in the form of an optical film. The cylindrical lens array 010 consists of multiple parallel-arranged cylindrical lens units, each corresponding to several pixel columns of the display panel 340. The optical function of the cylindrical lens array 010 is to refract and distribute the light emitted from the display panel 340 in a predetermined direction, allowing observers at different positions to receive light from different pixel columns, thereby creating stereoscopic vision. As can be seen from the above embodiments, the structural parameters of the cylindrical lens array 010 (such as pitch and tilt angle) directly affect the stereoscopic display effect and two-dimensional display quality of the display device 300. These parameters can be determined during the design phase of the display device 300 and precisely controlled during the manufacturing process.

[0032] In two-dimensional content display mode, the directional sampling effect of the lenticular lens array 010 can cause jagged artifacts in the edge areas. The image processing module 310 introduces the structural parameters of the lenticular lens array 010 as prior information and performs targeted anisotropic smoothing processing on the edge areas at the source end, thereby improving the display quality of two-dimensional content while keeping the physical structure of the display panel 340 unchanged.

[0033] based on Figure 3 The data stream of the display device 300 shown may include: an image providing module 320 compositing multiple windows or layers rendered by the application and operating system into a target resolution image to be displayed, and outputting it to an image processing module 310; the image processing module 310 performing edge detection, edge band generation, soft mask calculation, scale estimation, filter kernel parameter mapping, anisotropic filtering, and soft fusion on the image to be displayed based on the structural parameters of the cylindrical lens array 010, and outputting the anti-aliased target image to a display control submodule 331; and the display control submodule 331 performing image kernel operations through a computation shader and outputting pixel data to a color processing submodule 33. 2; The color processing submodule 332 converts the pixel data to the native color gamut and electro-optic characteristics of the display panel 340, and outputs the color-corrected pixel data to the link encoding submodule 333; The link encoding submodule 333 packages the pixel data into the timing and transmission format of the display link, and outputs the encoded pixel stream to the timing control submodule 334; The timing control submodule 334 unpacks the pixel stream into the in-panel timing, generates a panel driving signal to drive the display panel 340; The display panel 340 converts the panel driving signal into light output, and distributes the pixel emission direction to different viewing angles through the lenticular lens array 010, ultimately presenting a clear, jagged image.

[0034] Please see Figure 4 , Figure 4This is a schematic flowchart of an image processing method provided in an embodiment of this application. This image processing method can be applied to a display device with a cylindrical lens array, such as the one described above. Figure 3 The display device 300 shown is composed of... Figure 3 The image processing module 310 in the display device 300 shown executes this. For example... Figure 4 As shown, the image processing method may include the following steps S410 to S450.

[0035] Step S410: Perform edge detection processing on the image to be displayed to obtain edge information.

[0036] The image to be displayed is usually a full frame image, which can contain various content rendered by the application and operating system, such as user interface elements, text, video frames, or 3D scenes.

[0037] Edge detection processing is used to identify pixel locations in an image to be displayed where there are significant changes in brightness or color. These pixel locations typically correspond to the outlines of objects, the edges of text strokes, or the boundaries of graphic lines. In display devices with cylindrical lens arrays, these edge regions are high-risk areas for producing jagged artifacts.

[0038] Edge information is the output of edge detection processing, typically represented as an edge map. Each pixel value in the edge map indicates whether the location belongs to an edge and the strength of the edge. In this embodiment, the edge information may include not only the edge's location but also its direction and gradient features, to facilitate subsequent generation of edge bands and determination of filter kernel parameters.

[0039] Step S420: Determine the edge band and its soft mask based on the edge information; wherein the soft mask is used to indicate the fusion weight of each pixel within the edge band.

[0040] An edge band refers to a finite-width strip-shaped region constructed near an edge, covering a certain pixel range around the edge, and is a candidate region for anti-aliasing processing. This application uses edge bands to strictly limit subsequent filtering to the edge neighborhood, avoiding meaningless overall blurring of flat areas, textured detail areas, and non-critical structural areas, thereby reducing image sharpness loss.

[0041] A soft mask is a continuous weight distribution map generated within the edge band, used to indicate the fusion weight of each pixel within the edge band. The value of this fusion weight typically ranges from 0 to 1. If the original image is directly replaced with the filtered result within the edge band, abrupt changes can easily occur at the edge band boundaries, resulting in problems such as hazy edges, gray edges, and blurred outlines. The embodiments of this application achieve a smooth transition in processing intensity through a soft mask: a higher fusion weight is set in the central region of the edge band, resulting in a stronger filtering effect; a lower fusion weight is set in the boundary region of the edge band, gradually transitioning to the original image. This soft transition mechanism avoids the brightness abrupt changes and blurred outlines caused by hard replacement, helping to make the anti-aliasing effect visually more natural and the edge transition more continuous.

[0042] Step S430: Determine the filter kernel parameters of the anisotropic filter kernel based on the scale characteristics of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array.

[0043] A closed region refers to an internally connected region enclosed by continuous, closed boundaries in edge information. These regions represent independent structural units in an image, such as text strokes, graphic blocks, and icons. Scale features are quantitative descriptions of the geometric attributes of closed regions, which may include, but are not limited to, region area, total number of pixels, and stroke width. This application not only identifies edges but also further utilizes the internal regions enclosed by closed boundaries to estimate the local structural scale, extracting morphological features from these regions to adaptively select filter kernel parameters based on local region features. If a blur kernel is used for smaller regions, graying or blurring may occur; if a blur kernel is used for larger regions, jagged edges will remain, resulting in insufficient softening.

[0044] Structural parameters of a cylindrical lens array may include, for example, pitch and tilt angle. Pitch is the periodic distance between adjacent cylindrical lens units, which determines the spatial frequency of optical sampling; tilt angle is the angle of inclination of the cylindrical lens fringes relative to the horizontal direction of the display panel, which determines the direction of optical redistribution. These structural parameters affect the design of the filter core.

[0045] An anisotropic filter kernel is a direction-dependent smoothing kernel, where the smoothing intensity varies with direction. Unlike isotropic filter kernels (such as the standard Gaussian kernel), which smooth uniformly in all directions, anisotropic filter kernels can apply stronger smoothing in specific directions while maintaining weaker smoothing in others, thus better adapting to the directional sampling characteristics of cylindrical lens displays. Filter kernel parameters are a set of parameters describing the characteristics of anisotropic filter kernels, and may include, for example, scale parameters (controlling the smoothing range) and direction parameters (controlling the smoothing direction). By combining the scale features of the image content with the structural parameters of the cylindrical lens array, the optimal filter kernel parameters can be determined, achieving dual optimization of content adaptation and optical structure matching.

[0046] Specifically, the smoothing intensity of the anisotropic filter kernel in the direction perpendicular to the fringe of the cylindrical lens array is greater than that in the direction parallel to the fringe of the cylindrical lens array. This directional design is based on the physical mechanism of jagged edges produced by cylindrical lenses: along the direction perpendicular to the cylindrical lens fringe, the image is more prone to visual steps and jagged edges, so stronger smoothing should be applied; while along the fringe direction, it is often desirable to preserve the contour direction and texture extension, so weaker smoothing should be applied.

[0047] Step S440: According to the filter kernel parameters, perform filtering processing on the image to be displayed within the edge band to obtain the filtered image.

[0048] The filtering process involves convolving the image to be displayed using an anisotropic filtering kernel. Since the kernel parameters have been optimized based on the local image scale and lenticular structure, the filtering process effectively suppresses jagged edges while preserving the main outline and texture details of the image to the greatest extent possible. In this embodiment, the filtering process is performed only within the edge bands, leaving non-edge areas unchanged, thus ensuring that global sharpness is not compromised.

[0049] Step S450: According to the soft mask, perform pixel-by-pixel fusion processing on the filtered image and the image to be displayed within the edge band to obtain the target image for display.

[0050] Pixel-by-pixel fusion processing weights and synthesizes the filtered image and the original image to be displayed according to the fusion weights indicated by the soft mask. For each pixel within the edge band, the final output value can be the product of the pixel value of the filtered image and the fusion weight indicated by the soft mask, plus the product of the pixel value of the image to be displayed and (1 minus the fusion weight indicated by the soft mask). This fusion mechanism achieves a smooth, gradual transition within the edge band, avoiding artifacts such as fogging and broken strokes caused by hard thresholding.

[0051] In summary, the technical solution provided in this application involves edge detection processing of the image to be displayed to obtain edge information; determining the edge band and its soft mask based on the edge information; determining the filtering kernel parameters of the anisotropic filtering kernel based on the scale features of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array; filtering the image to be displayed within the edge band according to the filtering kernel parameters to obtain a filtered image; and performing pixel-by-pixel fusion processing on the filtered image and the image to be displayed within the edge band according to the soft mask to obtain the target image for display. Based on this, addressing the issues of edge jaggedness artifacts, color fringing, and flickering generated when displaying two-dimensional content using cylindrical lens 3D displays, this application embodiment, by processing only within the edge band, combining local image scale features and cylindrical lens structural parameters for anisotropic smoothing filtering, and achieving pixel-by-pixel fusion within the edge band using a soft mask, can suppress jaggedness while preserving the main image outline, stripe direction texture, and subjective sharpness as much as possible. This achieves adaptive and selective suppression of the local high-frequency structures most relevant to jaggedness without significantly sacrificing global sharpness, while also considering the directionality and scale dependence of cylindrical lens displays, effectively improving the display effect.

[0052] In some embodiments, step S410 may include: converting the image to be displayed to a linear luminance space; performing gradient calculations on the image to be displayed in the linear luminance space to obtain gradient information of each pixel in the image to be displayed; performing edge detection on the image to be displayed based on the gradient information to obtain an initial edge map; and performing connectivity repair on the initial edge map to obtain edge information.

[0053] Converting the image to be displayed to a linear luminance space establishes a computable linear domain, facilitating subsequent luminance calculations and gradient operations. Display devices typically store and transmit image data in non-linear spaces (such as sRGB), and performing calculations directly outside the linear domain can lead to luminance estimation errors and color distortion. However, after conversion to a linear luminance space, pixel values ​​exhibit a linear relationship with physical light intensity, ensuring the accuracy of subsequent edge detection and weight calculations.

[0054] Gradient calculation is used to calculate the rate of change of brightness of each pixel in the image to be displayed in the horizontal and vertical directions to obtain gradient information. The gradient information can include horizontal gradient components and vertical gradient components, which together describe the magnitude and direction of local brightness changes. In this embodiment, gradient calculation can be implemented using any first-order or second-order difference operator, such as the Sobel operator, Roberts operator, Scharr operator, Kirsch operator, or Laplacian operator. Specifically, the Sobel operator calculates the gradient using a weighted average, which has a certain smoothing effect on noise; the Roberts operator uses cross-difference, which is simple to calculate but sensitive to noise; the Scharr operator optimizes the weight coefficients of the Sobel operator, making it more accurate in edge direction estimation; the Kirsch operator detects edges through template matching in multiple directions; and the Laplacian operator is based on the second derivative, which is more accurate in edge localization but more sensitive to noise.

[0055] For example, in order to identify edge location, edge strength, and edge direction, the Sobel operator or an equivalent first-order difference operator can be used to calculate the horizontal gradient. and vertical gradient Based on this, the gradient magnitude can be obtained. and gradient direction The following formulas 1 and 2 are shown respectively, where the gradient direction is... It can represent the normal direction of the edge.

[0056] Formula 1: .

[0057] Formula 2: .

[0058] Edge detection can be performed based on gradient information to identify edge pixels by analyzing gradient magnitude and direction, thereby generating an initial edge map. Since the initial edge map is the direct output of edge detection, it may contain issues such as local breaks, small holes, jagged edges causing discrete gaps, or missing edges in weakly textured regions. Directly using the initial edge map could lead to incomplete region segmentation and unstable region scale estimation, further affecting the selection of filter kernel parameters. Therefore, this embodiment performs connectivity repair on the initial edge map to connect broken edges, fill gaps and holes, and finally output edge information.

[0059] In some embodiments, connectivity repair may include morphological closing and hole filling. Morphological closing is used to first dilate the initial edge map to connect adjacent broken edges, and then erosion is performed to restore edge thickness, thereby filling small gaps. Hole filling is used to fill internal voids in the closed region, forming a complete closed region. After connectivity repair, the boundary structure in the edge information is relatively coherent, and it can stably distinguish between the "boundary" and the "internal region enclosed by the boundary," facilitating subsequent connected component segmentation and scale estimation.

[0060] In summary, the technical solution provided in this application converts the image to be displayed to a linear brightness space for gradient calculation and edge detection, and introduces a connectivity repair mechanism to handle edge breakage and hole problems. This enables the acquisition of more accurate and stable edge information, thereby improving the reliability of subsequent edge band generation, scale estimation and filtering processing, and further improving the anti-aliasing effect.

[0061] In some embodiments, step S420 may include: performing morphological dilation processing based on edge information to obtain an edge band; determining the fusion weight of each pixel in the edge band based on the distance between each pixel in the edge band and the boundary of the edge band to generate a soft mask; wherein the fusion weight is positively correlated with the distance.

[0062] Morphological dilation is an expansion operation based on edge information, extending edge pixels to their neighborhoods by a certain radius to generate an edge band covering the surrounding edge. The dilation radius is a key parameter controlling the width of the edge band: a larger dilation radius allows for a wider processed edge neighborhood and a larger processing range; a smaller dilation radius results in more localized and conservative processing. In some embodiments, the dilation radius can be positively correlated with the pitch of the cylindrical lens array, because a larger pitch results in a longer spatial period for optical sampling and a wider range of edge jaggedness, thus requiring a wider edge band for coverage.

[0063] By performing a distance transformation on the edge band, the distance between each pixel within the edge band and the edge band boundary is calculated, thereby determining the fusion weight of each pixel to generate a soft mask. This distance reflects the pixel's position relative to the edge band center; pixels with larger distances are closer to the edge band center, and pixels with smaller distances are closer to the edge band boundary. There is a positive correlation between the fusion weight and the distance. In this embodiment, the distance can be normalized to the range of 0 to 1 to obtain the fusion weight, as shown in Formula 3 below.

[0064] Formula 3: .

[0065] in, For pixels The fusion weight, For pixels The distance between the edge zone and the boundary. This is the thickness parameter for soft transition. This is an empirical parameter controlling the soft mask transition speed, typically ranging from 1 to 10 pixels, for example, set to 2 to 5 pixels. In some embodiments, There is a positive correlation between the pitch and the size of the cylindrical lens array. This is because a larger pitch results in a longer optical sampling period, requiring a thicker, softer transition at the edges to avoid abrupt changes. The value can be set to be larger.

[0066] By leveraging the positive correlation between fusion weights and distance, pixels closer to the edge center (larger distance) receive higher fusion weights, contributing more to the filtered image and resulting in a stronger smoothing effect; conversely, pixels closer to the edge boundary (smaller distance) receive lower fusion weights, contributing more to the original image and preserving its original appearance. This gradient mechanism ensures a smooth decay of processing intensity from the edge center to the edge boundary, avoiding abrupt brightness changes and graying of contours caused by hard replacement.

[0067] In summary, the technical solution provided in this application generates an edge band associated with the cylindrical lens pitch through morphological dilation processing, and generates a continuously gradient soft mask based on distance transformation. This achieves fine control over "where the edge band is processed and how the processing intensity changes smoothly" and effectively suppresses jagged edges while maximizing the protection of the clarity of non-edge areas.

[0068] In some embodiments, step S430 may include steps S431 to S433.

[0069] Step S431: Perform connected component segmentation based on edge information to obtain closed regions.

[0070] Connected component segmentation is used to identify and separate the internal regions enclosed by closed boundaries in edge information, obtaining at least one closed region. Each closed region can represent an independent structural unit in the image, such as a character stroke, a graphic element, or an icon. Connected component segmentation can extract all closed boundaries from the edge information based on pixel connectivity criteria (such as 4-connectivity or 8-connectivity) and label the internal region enclosed by each boundary.

[0071] Step S432: Determine the basic scale parameters of the anisotropic filter kernel based on the scale characteristics of the closed region.

[0072] Scale features are quantitative descriptions of the geometric properties of a closed region. In some embodiments, the scale feature can be the total number of pixels within the closed region, i.e., the equivalent pixel count. The equivalent pixel count reflects the size of the closed region: the larger the equivalent pixel count, the larger the area of ​​the closed region; the smaller the equivalent pixel count, the smaller the area of ​​the closed region. Based on the scale features of the closed region, the fundamental scale parameters of the anisotropic filter kernel can be adaptively selected.

[0073] In some embodiments, step S432 may include: determining the scale level of the closed region based on the scale characteristics of the closed region; determining the basic scale parameters of the anisotropic filter kernel based on the scale level and scale parameter mapping data; wherein the scale parameter mapping data includes at least one set of correspondences between scale levels and basic scale parameters.

[0074] Scale levels are the result of classifying closed regions according to their size. This application does not limit the specific settings of scale levels in its embodiments; in practical applications, they can be flexibly set according to requirements. For example, scale levels may include three levels: small region, medium region, and large region. Taking the scale feature including the equivalent number of pixels as an example, a small region can correspond to an equivalent number of pixels. Less than the first threshold The closed region, the middle region can correspond to the equivalent number of pixels. Between the first threshold Second threshold A closed region between these areas can correspond to an equivalent number of pixels. Greater than or equal to the second threshold The closed region. The specific values ​​of the first and second thresholds can be arbitrarily and flexibly set according to the actual application scenario and the characteristics of the display device.

[0075] Based on the scale level and scale parameter mapping data, the fundamental scale parameters of the anisotropic filter kernel can be determined. The scale parameter mapping data includes at least one set of correspondences between scale levels and fundamental scale parameters. Taking a scale level comprising small, medium, and large domains as an example, the small domain is mapped to the fundamental scale parameters. The mid-domain mapping is based on the scale parameter. Large-area mapping is based on the scale parameter. As shown in Formula 4 below, where < < .

[0076] Formula 4: .

[0077] Based on this, a larger scale level of closed region corresponds to a larger filter kernel scale parameter. This is because large regions require a larger smoothing range to effectively suppress jagged edges, while using a large kernel in small regions can easily result in graying or blurring.

[0078] Step S433: Determine the filter kernel parameters based on the basic scale parameters and the structural parameters of the cylindrical lens array.

[0079] The structural parameters of the cylindrical lens array may include pitch and tilt angle, and the filter kernel parameters of the anisotropic filter kernel may include target scale parameters and target orientation parameters. In some embodiments, step S433 above may include: correcting the basic scale parameters according to the pitch of the cylindrical lens array to obtain the target scale parameters; and determining the target orientation parameters of the anisotropic filter kernel according to the tilt angle of the cylindrical lens array.

[0080] Pitch, the periodic distance between adjacent cylindrical lens units, determines the spatial frequency of optical sampling. Pitch is typically positively correlated with filter intensity: a larger pitch results in a longer spatial period for optical sampling, leading to more pronounced spatial widening of edge jaggedness and requiring a larger filter scale to cover it. Therefore, by modifying the base scale parameters based on the pitch, and further considering fine-tuning of optical structural characteristics on top of the image content scale, a target scale parameter that better matches the actual display characteristics can be obtained. In some embodiments, the pitch of the cylindrical lens array is positively correlated with the target scale parameter of the anisotropic filter kernel. For example, the base scale parameter can be linearly scaled according to the ratio between the pitch of the cylindrical lens array and a preset reference pitch to obtain the target scale parameter.

[0081] The tilt angle, or inclination angle, of the cylindrical lens fringes relative to the horizontal direction of the display panel determines the direction of optical redistribution. The fringe direction of the cylindrical lens array can be determined based on the tilt angle; the fringe direction is parallel to the direction in which the cylindrical lens units extend. The target orientation parameter of the anisotropic filter core is determined by the fringe direction to guide the orientation of the anisotropic filter core in space.

[0082] In some embodiments, the anisotropic filter kernel can be an elliptical Gaussian kernel, and the target direction parameters can include the major axis direction and the minor axis direction. The above-mentioned determination of the target direction parameters of the anisotropic filter kernel based on the tilt angle of the cylindrical lens array can include: determining the fringe direction of the cylindrical lens array based on the tilt angle of the cylindrical lens array; setting the major axis direction to be perpendicular to the fringe direction of the cylindrical lens array, and setting the minor axis direction to be parallel to the fringe direction of the cylindrical lens array.

[0083] An elliptic Gaussian kernel is a variation of a two-dimensional Gaussian function, whose contour lines are ellipses rather than circles. The major axis direction is where the smoothing intensity of the elliptic Gaussian kernel is strongest, and the minor axis direction is where the smoothing intensity is weakest. Setting the major axis direction perpendicular to the fringe direction of the cylindrical lens array allows for the strongest smoothing in the direction most prone to jagged edges; setting the minor axis direction parallel to the fringe direction of the cylindrical lens array allows for the weakest smoothing in the direction where contour preservation is required. This directional setting achieves the goal of "suppressing jagged edges in the normal direction and preserving contours in the tangential direction," taking into account the directionality of cylindrical lens displays.

[0084] For example, an elliptical Gaussian kernel in a rotating coordinate system It can be represented by the following formula 5. Wherein, The direction is the major axis of the elliptical Gaussian kernel, perpendicular to the fringe direction of the cylindrical lens array; The major axis scaling parameter of the elliptic Gaussian kernel (as mentioned above) , , All of these can be implemented as major axis scale parameters. The direction is the minor axis of the elliptical Gaussian kernel, parallel to the fringe direction of the cylindrical lens array; These are the minor axis scaling parameters of the elliptic Gaussian kernel, and they satisfy a certain ratio. This ratio is usually between 0 and 1, for example, 0.5.

[0085] Formula 5: .

[0086] It should be understood that the anisotropic filtering kernel in the embodiments of this application is not limited to an elliptic Gaussian kernel, but can also be implemented as a guided filtering kernel or a bilateral filtering kernel, etc., and can be flexibly set according to requirements in practical applications. The guided filtering kernel uses the luminance component of the image to be displayed as the guiding image, and adopts a first local window scale in the fringe direction perpendicular to the cylindrical lens array and a second local window scale in the fringe direction parallel to the cylindrical lens array. The first local window scale is larger than the second local window scale, so that the smoothing intensity of the guided filtering kernel in the fringe direction is greater than the smoothing intensity in the fringe direction parallel to the fringe. The spatial domain standard deviation of the bilateral filtering kernel has a first value in the fringe direction perpendicular to the cylindrical lens array and a second value in the fringe direction parallel to the cylindrical lens array. The standard deviation of the range is adaptively adjusted according to the local gradient magnitude to achieve direction-dependent anisotropic smoothing.

[0087] In summary, the technical solution provided in this application achieves dual optimization of image content adaptation and optical structure matching by extracting closed regions through connected component segmentation, performing scale-level mapping based on the scale characteristics of the closed regions, and determining the filter kernel parameters by combining the pitch and tilt angle of the cylindrical lens array. This effectively suppresses jagged edges while maximizing the preservation of image details and subjective sharpness.

[0088] In some embodiments, the image processing method described above may further include: determining the angle difference between the edge tangential direction in the edge information and the stripe direction of the cylindrical lens array; generating a direction gating weight based on the angle difference; and dynamically adjusting the smoothing intensity of the anisotropic filter kernel during the filtering process based on the direction gating weight.

[0089] The edge tangential direction is the tangential direction of the local edge orientation and is perpendicular to the edge normal direction. The edge tangential direction is a key criterion for determining whether an edge is susceptible to the influence of cylindrical lens fringes. When the edge tangential direction is approximately consistent with the fringe direction of the cylindrical lens array, the direction crossing the edge is perpendicular to the fringe direction of the cylindrical lens array. The cylindrical lens array exhibits significant optical redistribution capability in this direction, making it most prone to producing noticeable serrations. When the edge tangential direction has a large angle with the fringe direction of the cylindrical lens array, especially when it is close to perpendicular, the direction crossing the edge is approximately parallel to the fringe direction of the cylindrical lens array. The optical redistribution capability of the cylindrical lens array in this direction is weaker, and the risk of serrations is significantly reduced.

[0090] The included angle difference is the angular difference between the edge tangential direction and the fringe direction of the cylindrical lens array. It is typically taken as the minimum included angle (considering directional periodicity). A smaller included angle difference indicates that the edge tangential direction is more parallel to the fringe direction of the cylindrical lens array, and the edge normal direction is more perpendicular to the fringe direction, resulting in a higher risk of jagged edges. Conversely, a larger included angle difference indicates that the edge tangential direction is more perpendicular to the fringe direction, and the edge normal direction is more parallel to the fringe direction, resulting in a lower risk of jagged edges. For example, the included angle difference... It can be shown in Formula 6 below. The edge tangential direction can be determined based on the gradient direction mentioned above, i.e. ; The direction of the stripes in the cylindrical lens array.

[0091] Formula 6: .

[0092] The directional gating weight is a dynamic adjustment factor generated based on the angle difference. In this embodiment, the directional gating can be calculated based on Gaussian attenuation, meaning that the directional gating weight has a negative exponential relationship with the square of the angle difference. For example, the directional gating weight can be shown in Formula 7 below.

[0093] Formula 7: .

[0094] in, For direction gating weights, This refers to the directional bandwidth parameter. The directional gating weight is used to control the allowable angle range, which is related to the tilt angle of the cylindrical lens array and is typically between ±90 degrees, for example, ±45 degrees. The reference direction is perpendicular to the tilt angle of the cylindrical lens array. When the angle difference is small (higher risk of aliasing), the directional gating weight is close to 1, applying strong smoothing; when the angle difference is large (lower risk of aliasing), the directional gating weight decreases, and the smoothing intensity is suppressed; when the angle difference is close to 90 degrees, the directional gating weight approaches 0, and smoothing is turned off.

[0095] During the filtering execution phase, for each pixel within the edge band, the angle difference between its edge tangential direction and the stripe direction can be calculated. A corresponding direction-gating weight is then generated, and this weight is multiplied by the baseline smoothing intensity of the filter kernel to obtain the actual smoothing intensity. This dynamic adjustment ensures that only the parts where the local edge tangential direction is approximately aligned with the cylindrical lens direction are softened, while the softening intensity is reduced or even turned off for edges with mismatched directions. This avoids over-smoothing of unnecessary edges and further preserves details and sharpness in other directions of the image.

[0096] In summary, the technical solution provided in this application, by introducing a directional gating mechanism, dynamically adjusts the smoothing intensity during the filtering process based on the degree of matching between the local edge direction and the stripe direction of the cylindrical lens array, thereby further improving edge selectivity. While effectively suppressing cylindrical lens-related jaggedness, it maximizes the protection of structural details and texture sharpness in other directions of the image.

[0097] The image processing method provided in the embodiments of this application will be described below with several examples.

[0098] Please see Figure 5 , Figure 5 This is a schematic flowchart of another image processing method provided in an embodiment of this application. This image processing method can be applied to a display device with a cylindrical lens array, such as the one described above. Figure 3 The display device 300 shown is composed of... Figure 3 The image processing module 310 in the display device 300 shown executes this. For example... Figure 5 As shown, the image processing method may include the following steps S501 to S510.

[0099] Step 501: Input preprocessing, bit depth normalization. Receive the entire frame of the image to be displayed, which can be rendered and composited from multiple windows or layers. Unify the image to be displayed into a computable domain, and normalize the bit depth and numerical range of the pixel data to the range of 0 to 1 for easier subsequent computation.

[0100] Step 502: Linearization and Color Preparation. The image to be displayed is converted from a non-linear color space (such as sRGB) to a linear luminance space to eliminate the influence of the non-linear color space on luminance calculations. The linear luminance component Y is calculated, for example, using a weighted method of Y = 0.2126R + 0.7152G + 0.0722B. Subsequent processing performs calculations in the luminance domain while preserving the original RGB data for use in the final restoration and blending stages, preventing channel drift and jagged edges caused by independent processing of the RGB channels.

[0101] Step 503: Gradient and Structure Tensor Estimation. To identify edge location, edge strength, and edge direction, the Sobel operator or an equivalent first-order difference operator can be used to calculate the horizontal gradient. and vertical gradient Based on this, the gradient magnitude can be obtained. and gradient direction This processing is not limited to the Sobel operator; first-order or second-order difference operators such as the Roberts operator, Scharr operator, Kirsch operator, and Laplacian operator can also be used.

[0102] Step 504: Edge Band and Soft Mask Generation. Construct edge bands of finite width near the edges as candidate regions for anti-aliasing. Perform a distance transform on the edge bands, calculate the distance between each pixel within the edge band and its outer boundary, and normalize this distance to a range of 0 to 1 to generate the soft mask. The edge band determines "where can be processed", and the soft mask determines "how the processing intensity changes smoothly", avoiding abrupt brightness changes and graying of outlines caused by hard replacement.

[0103] Step 505: Closed Region Segmentation and Feature Extraction. Connected region segmentation is performed on the internal connected regions enclosed by the boundaries. Morphological features are extracted from these regions to estimate the local structural scale. This process restores potentially broken or unclosed boundaries into relatively coherent boundary structures, enabling a stable distinction between the "boundary" and the "internal region enclosed by the boundary" in subsequent steps.

[0104] Step 506: Scale-Kernel Parameter Mapping. The closed region is divided into corresponding scale levels, such as small, medium, and large regions, according to scale features (e.g., equivalent pixel count), and mapped to the corresponding filter kernel scale parameters: small regions correspond to smaller base scale parameters. The medium domain corresponds to the basic scale parameter of the medium domain. Larger domains correspond to larger fundamental scale parameters. Based on this, we can achieve adaptive processing of "small cores for small domains and large cores for large domains" to avoid over-smearing or under-suppression problems caused by a unified core.

[0105] Step 507: Lens Parameter Correction Anisotropic Filtering. The pitch and tilt angle of the cylindrical lens array are introduced as prior information to dynamically adjust the anisotropic direction and intensity of the filter kernel. An elliptical Gaussian kernel can be used as the anisotropic filter kernel, with the major axis set perpendicular to the fringe direction of the cylindrical lens array and the minor axis set parallel to the fringe direction. A larger pitch results in a wider edge band and a stronger filter intensity. The filter kernel supports separable implementation and rotational domain convolution, facilitating low-overhead deployment in GPU / ISP / display pipelines.

[0106] Step 508: Direction Gating. Calculate the angle difference between the local edge tangential direction and the fringe direction of the cylindrical lens array. Generate Gaussian attenuation direction gating weights based on the angle difference. The softening is enhanced only on the local edges whose tangential direction is approximately consistent with the direction of the cylindrical lens, while the softening intensity is reduced or even turned off on edges whose directions do not match, thus avoiding over-smoothing of edges that do not need to be processed.

[0107] Step 509: Luminosity-driven selective fusion. The filtered result is fused with the original image to be displayed using an energy-consistent method within the edge band using a soft mask. Fusion weights are estimated in the luminance domain, but the same fusion weight is used for all three RGB channels to suppress color fringing and channel drift.

[0108] Step 510: Output and Encoding. The processing result is sent back from the linear domain to the display link standard format and then transmitted by the color and physical layers to ensure that the algorithm's benefits are faithfully and accurately presented on the terminal display.

[0109] Based on this, the image processing method in the embodiments of this application can be summarized by the following formula 8.

[0110] Formula 8: .

[0111] in, For the output target image in pixels Pixel value at; For the image to be displayed in pixels Pixel value at; For pixels The anisotropic filter kernel corresponding to the scale level of the closed region. .

[0112] In summary, the innovations of this application's embodiments include: using edge detection as the entry point, processing is only performed in high-risk jagged areas (the edge band after expansion / closure and its inner closed area), with zero modification to non-edge areas, maintaining global clarity; connected component segmentation and feature estimation such as area / stroke width are performed on the inner region after edge closure, and the filtering intensity and kernel size are mapped according to scale hierarchy, where small kernels are used for small regions and large kernels are used for large regions to avoid over-smearing / under-suppression caused by a uniform kernel; one of elliptical Gaussian / guided filtering / bilateral filtering is applied in the edge normal and tangential coordinate systems to meet the goal of "suppressing jaggedness in the normal direction and preserving the contour in the tangential direction," while also taking into account lenticular lens display. The filter is directional; a soft weighting method combining distance field and gradient magnitude is used to construct a transition band, and the output is a weighted fusion of the original image and the filtered image, avoiding fog edges / broken strokes caused by hard threshold replacement; weights are estimated in the brightness (or linear light domain), but the same fusion weight is used for the three RGB channels, which suppresses color edge jaggedness and channel drift from a mechanism perspective; the pitch and tilt angle of the cylindrical lens array are introduced as prior information to dynamically adjust the direction and intensity of the filter anisotropy, achieving robust suppression related to screen position; the filter kernel supports separable implementation and rotation domain convolution, which helps to deploy with low overhead in GPU / ISP / display pipeline; the parameter mapping is continuously adjustable, which facilitates calibration and automated parameter tuning.

[0113] Please see Figure 6 , Figure 6 This is a schematic diagram illustrating the effect of image processing provided in an embodiment of this application. For example... Figure 6 As shown, the original image is a black curve; edge band generation creates red band-shaped areas on both sides of the curve through morphological dilation; the soft mask uses pseudo-color to represent the distance distribution between each pixel within the edge band and the edge band boundary, with brighter colors indicating greater distances and higher fusion weights; the orientation gating weights display a dynamic adjustment factor calculated based on the angle difference between the edge tangent and the direction of the cylindrical lens fringe, with higher weights (yellowish color) in areas where the edge tangent is approximately parallel to the fringe direction, and lower weights (blueish color) in areas where they are approximately perpendicular; finally, the soft mask and the orientation gating weights are multiplied pixel by pixel. Figure 6 As shown, the directional gating weights locally modulate the soft mask, retaining higher weights in high-risk areas of aliasing that require focused suppression, while reducing processing intensity at edges in other directions, thus achieving selective and directional fine-grained smoothing control.

[0114] Please see Figure 7 , Figure 7 This is a schematic diagram illustrating the effect of another image processing method provided in an embodiment of this application. Figure 7 The images are taken from real-life photos taken with a mobile phone to simulate the effect of human eye observation. For example... Figure 7As shown in (1) above, in the image on the left before optimization, obvious jagged artifacts appear at the edges of the vertical stripes, manifested as step-like undulations and uneven local brightness at the edges; in the image on the right after optimization based on the technical solution provided in the embodiments of this application, the edges of the vertical stripes become smooth and continuous, the jaggedness is significantly reduced, and the edge transition is more uniform. Figure 7 As shown in (2) above, the image on the left before optimization exhibits obvious jagged edges and graying at the intersections of horizontal and vertical lines, as well as at the edges of the lines, resulting in decreased line clarity. In the image on the right after optimization based on the technical solution provided in this application, the edges of the intersecting lines are smooth, the line sharpness is maintained, and there is no excessive blurring or hazy edge problem. It can be seen that the technical solution provided in this application can effectively maintain the main outline and subjective clarity of the image while suppressing jagged edges in the cylindrical lens display, proving the effectiveness of the edge anti-aliasing processing in this application.

[0115] The above provides a detailed description of an image processing method and display device provided by the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An image processing method, characterized in that, Applied to a display device having a cylindrical lens array, the method includes: Edge detection processing is performed on the image to be displayed to obtain edge information; Based on the edge information, an edge band and a soft mask of the edge band are determined; wherein, the soft mask is used to indicate the fusion weight of each pixel within the edge band; Based on the scale characteristics of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array, the filter kernel parameters of the anisotropic filter kernel are determined; wherein, the smoothing intensity of the anisotropic filter kernel in the fringe direction perpendicular to the cylindrical lens array is greater than the smoothing intensity of the anisotropic filter kernel in the fringe direction parallel to the cylindrical lens array. According to the filtering kernel parameters, the image to be displayed is filtered within the edge band to obtain a filtered image; According to the soft mask, the filtered image and the image to be displayed are fused pixel by pixel within the edge band to obtain the target image for display.

2. The method according to claim 1, characterized in that, The step of determining the edge band and the soft mask of the edge band based on the edge information includes: The edge band is obtained by performing morphological dilation processing based on the edge information; The fusion weights of each pixel within the edge band are determined based on the distance between each pixel within the edge band and the boundary of the edge band, in order to generate the soft mask; wherein the fusion weights are positively correlated with the distance.

3. The method according to claim 1, characterized in that, The step of determining the filter kernel parameters of the anisotropic filter kernel based on the scale characteristics of the closed region corresponding to the edge information and the structural parameters of the cylindrical lens array includes: The closed region is obtained by segmenting the connected components based on the edge information; Based on the scale characteristics of the closed region, the basic scale parameters of the anisotropic filter kernel are determined; The filter kernel parameters are determined based on the basic scale parameters and the structural parameters of the cylindrical lens array.

4. The method according to claim 3, characterized in that, The step of determining the basic scale parameters of the anisotropic filter kernel based on the scale characteristics of the closed region includes: The scale level of the closed region is determined based on its scale characteristics. Based on the scale level and scale parameter mapping data, the basic scale parameters of the anisotropic filter kernel are determined; wherein, the scale parameter mapping data includes at least one set of correspondences between the scale level and the basic scale parameters.

5. The method according to claim 3, characterized in that, Determining the filter kernel parameters based on the basic scale parameters and the structural parameters of the cylindrical lens array includes: Based on the pitch of the cylindrical lens array, the basic scale parameters are corrected to obtain the target scale parameters; The target direction parameters of the anisotropic filter core are determined based on the tilt angle of the cylindrical lens array. The structural parameters of the cylindrical lens array include the pitch and the tilt angle, and the filter kernel parameters include the target scale parameter and the target direction parameter.

6. The method according to claim 5, characterized in that, The anisotropic filter kernel is an elliptic Gaussian kernel, and the target direction parameters include the major axis direction and the minor axis direction; determining the target direction parameters of the anisotropic filter kernel based on the tilt angle of the cylindrical lens array includes: The fringe direction of the cylindrical lens array is determined based on the tilt angle of the cylindrical lens array. The major axis direction is set to be perpendicular to the stripe direction of the cylindrical lens array, and the minor axis direction is set to be parallel to the stripe direction of the cylindrical lens array.

7. The method according to claim 1, characterized in that, The method further includes: Determine the angle difference between the edge tangential direction in the edge information and the stripe direction of the cylindrical lens array; Based on the angle difference, a direction gating weight is generated; Based on the directional gating weights, the smoothing intensity of the anisotropic filter kernel is dynamically adjusted during the filtering process.

8. The method according to claim 1, characterized in that, The edge detection processing of the image to be displayed to obtain edge information includes: Convert the image to be displayed to a linear brightness space; Gradient calculations are performed on the image to be displayed in the linear brightness space to obtain the gradient information of each pixel in the image to be displayed. Based on the gradient information, edge detection is performed on the image to be displayed to obtain an initial edge map; The initial edge map is connected to obtain the edge information.

9. A display device, characterized in that, The display device has a cylindrical lens array and includes an image processing module for performing the image processing method as described in any one of claims 1 to 8.

10. The display device according to claim 9, characterized in that, The display device further includes: An image providing module, connected to the image processing module, is used to provide the image to be displayed to the image processing module; The display processing module is connected to the image processing module and is used to perform display processing on the target image output by the image processing module to obtain panel driving signals. The display panel, connected to the display processing module, includes the lenticular lens array, which is used to convert the panel driving signal into light output and distribute the pixel emission direction to different viewing angles through the lenticular lens array.

11. The display device according to claim 10, characterized in that, The image providing module is also used to: combine multiple windows or layers into a whole frame image of the target resolution to obtain the image to be displayed.

12. The display device according to claim 10, characterized in that, The display processing module includes: The display control submodule is connected to the image processing module and is used to perform image kernel operations of the image processing module through a compute shader, and output pixel data based on the target image; A color processing submodule, connected to the display control submodule, is used to convert the pixel data to the native color gamut and electro-optical characteristics of the display panel to obtain the color-corrected pixel data; The link encoding submodule, connected to the color processing submodule, is used to package the color-corrected pixel data into the timing and transmission format of the display link to obtain a pixel stream; The timing control submodule, connected to the link encoding submodule and the display panel, is used to unpack the pixel stream into in-panel timing and generate the panel drive signal based on the in-panel timing.

Images