Image chromatic aberration correction method, device and equipment based on 5D lookup table

Abstract

The invention discloses an image chromatic aberration correction method, device and equipment based on a 5D lookup table, and relates to the technical field of image processing methods. Comprising the following steps: mapping an image from an RGB space to a chromatic aberration space, decoupling and separating artifacts of the image in the chromatic aberration space into an independent chromatic aberration channel, a brightness channel and an orthogonal channel; in a brightness channel, through a direction perception 5D lookup table, geometric gradient information extracted from a color difference channel is used as a part of a lookup index to realize directional guidance of brightness restoration, and the color difference channel performs global intensity suppression through a 1D lookup table; and re-fusing the brightness channel data and the chromatic aberration channel data which are subjected to double-path correction with the unmodified orthogonal channel data, converting the image from a chromatic aberration space to a standard RGB color space, and finally outputting a high-quality image. The method is high in processing speed and good in purple edge removing effect, and colors and details of normal image content are not damaged.

Description

Technical Field

[0001] This invention relates to the field of image processing methods, and in particular to an image color difference correction method, apparatus, and device based on a 5D lookup table. Background Technology

[0002] Optical noise such as "purple fringing" or "purple halo" has long existed in digital imaging technology. Its physical cause is longitudinal chromatic aberration (LCA) of the lens, meaning that light of different wavelengths cannot be focused on the same plane, resulting in chromatic dispersion at high-contrast edges, forming purple or magenta outlines, severely degrading image sharpness and realism. Current industry solutions to this problem are mainly based on traditional manual feature extraction methods, all of which have insurmountable limitations.

[0003] The first type is the hardware solution, which physically corrects chromatic aberration by designing extremely complex and expensive optical systems such as apochromatic (APO) lenses. Although this solution is effective, its high cost and size make it completely unsuitable for mainstream consumer electronics products such as smartphones.

[0004] The second category is traditional computational methods. These methods rely on manually designed image processing workflows, such as first locating high-contrast areas through edge detection, then identifying color differences using fixed color thresholds, and finally performing desaturation operations. However, this logic based on fixed rules has fundamental flaws. Not only does it often lead to "false positives" and color distortion because it cannot accurately distinguish between artifacts and real purple objects in the image, but also, due to the diverse forms of color differences, fixed processing methods often fail to remove them completely, leaving residual color casts. Furthermore, its effectiveness heavily depends on manually adjusted parameters, resulting in extremely poor versatility.

[0005] The third category is the general-purpose deep learning restoration models that have emerged in recent years. Although these models perform well in tasks such as denoising and deblurring, they are equally inadequate when dealing with purple fringing. The fundamental reason is that these general-purpose models lack specific sensing biases for the physical causes of purple fringing (i.e., spatial misalignment of color channels), and cannot perform accurate physical modeling. Existing learnable lookup table (LUT) techniques mainly focus on global color enhancement rather than local, spatially sensitive artifact correction. Summary of the Invention

[0006] The technical problem to be solved by the present invention is how to provide an image color difference correction method based on a 5D lookup table that has a fast processing speed, good purple edge removal effect, and does not damage the color and details of normal image content.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an image color difference correction method based on a 5D lookup table, comprising the following steps: Color perception transformation: Maps an image from RGB space to a color difference space. In the color difference space, image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. Direction-aware correction: In the luminance channel, a directional 5D lookup table is used to guide the luminance restoration by using the geometric gradient information extracted from the chrominance channel as part of the lookup index. The chrominance channel itself uses a 1D lookup table for global intensity suppression. Fusion and Inverse Mapping: The luminance and chrominance channel data, after dual-path correction, are re-fused with the unmodified orthogonal channel data. Then, through the inverse mapping operation of the transformation module, the image is converted from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

[0008] This invention also discloses an image color difference correction device based on a 5D lookup table. The device employs the aforementioned image color difference correction method based on a 5D lookup table, and the device includes: Color perception transformation module: used to map the image from RGB space to a color difference space. In the color difference space, the image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. Direction-aware correction module: In the luminance channel, it uses a direction-aware 5D lookup table and the geometric gradient information extracted from the chromatic difference channel as part of the lookup index to achieve directional guidance for luminance repair. The chromatic difference channel itself uses a 1D lookup table for global intensity suppression. The fusion and inverse mapping module is used to re-fuse the dual-corrected luminance and chrominance channel data with the unmodified orthogonal channel data. Through the inverse mapping operation of the transformation module, it converts the image from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

[0009] The present invention also discloses a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the image color difference correction method based on a 5D lookup table.

[0010] The beneficial effects of adopting the above technical solution are as follows: the method can learn a color space that adapts to the image content through chromaticity perception transformation, thereby intelligently decoupling and separating the chromatic aberration artifact information caused by the longitudinal chromatic aberration (LCA) of the lens from the complex image background into an independent chromatic aberration channel, so that the correction process only targets the artifact itself, fundamentally ensuring that the color and details of the normal image content are not damaged.

[0011] Secondly, by using a Direction-Aware 5D Look-Up Table, this invention utilizes the geometric gradient information contained in the color difference channel separated in the previous step to guide the repair of the luminance channel. This directly responds to the physical nature of the color channel spatial misalignment caused by LCA, achieving directional, physically driven, and precise correction, resulting in sharper and cleaner repair results.

[0012] Finally, this invention solves the problems of cost in hardware solutions and efficiency in general models. Its lightweight network design (only 0.13M parameters) can process 2K and even 4K ultra-high-definition images in real time on resource-constrained devices such as smartphones, and has the advantages of high efficiency and high speed. Attached Figure Description

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0014] Figure 1 This is the main flowchart of the method described in the embodiments of the present invention; Figure 2 This is a schematic block diagram of the device described in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the processing of the device described in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] Example 1: Overall, such as Figure 1 As shown in the figure, an embodiment of the present invention discloses an image color difference correction method based on a 5D lookup table, the method comprising the following steps: S1, Color Perception Transformation: Maps the image from RGB space to a color difference space. In the color difference space, the image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. S2, Direction-aware correction: In the luminance channel, a directional guidance for luminance restoration is achieved by using a directional 5D lookup table and the geometric gradient information extracted from the chrominance channel as part of the lookup index. The chrominance channel itself performs global intensity suppression through a 1D lookup table. S3, Fusion and Inverse Mapping: The luminance channel data and chrominance channel data after dual-path correction are re-fused with the unmodified orthogonal channel data. Then, through the inverse mapping operation of the transformation module, the image is converted from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

[0018] Example 2 like Figure 2 As shown, the present invention also discloses an image color difference correction device based on a 5D lookup table. The device employs the aforementioned image color difference correction method based on a 5D lookup table, and the device includes: Color perception transformation module 101: used to map an image from RGB space to a color difference space. In the color difference space, the image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. Direction-aware correction module 102: In the luminance channel, it uses a direction-aware 5D lookup table and the geometric gradient information extracted from the chromatic difference channel as part of the lookup index to achieve directional guidance for luminance repair. The chromatic difference channel itself uses a 1D lookup table for global intensity suppression. The fusion and inverse mapping module 103 is used to re-fuse the dual-corrected luminance channel data and chrominance channel data with the unmodified orthogonal channel data, and through the inverse mapping operation of the transformation module, convert the image from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

[0019] like Figure 3 The overall processing flow of the device is shown below: Given an input image containing chromatic aberration noise The core idea of ​​the method is not to process directly in the RGB space, but to first use a learnable chromaticity-aware transformation module to map the image from the complex RGB space to a new, problem-oriented chromatic difference space. In this unique space, artifacts are intelligently decoupled and separated into an independent chromatic difference channel, while the main structure and brightness information of the image are retained in the brightness channel.

[0020] Next, the core direction-aware correction module is entered, which is a dual-path parallel processing mechanism: the correction of the luminance channel is performed by an innovative direction-aware 5D-LUT, which is unique in that it uses the geometric gradient information extracted from the chrominance channel as part of the lookup index, thereby achieving precise directional guidance for luminance repair; at the same time, the purple edge channel itself is subjected to global intensity suppression through a simpler 1D-LUT.

[0021] Finally, the channel data after dual-path correction is re-fused and, through the inverse mapping operation of the transformation module, is accurately converted from the color difference space back to the standard RGB color space, ultimately producing and outputting a visually clean, clear, and high-quality image that successfully removes color difference artifacts. .

[0022] The device first addresses the drawbacks of traditional methods, such as accidental and residual artifacts, by designing a Color Aware Transform (CA-CT) module. This module learns a color space that adapts to the image content, intelligently decoupling and separating the chromatic aberration artifact information caused by LCA from the complex image background into an independent chromatic aberration channel. This ensures that the correction process targets only the artifacts themselves, fundamentally protecting the color and detail of the normal image content. Secondly, it addresses the lack of specificity in general models.

[0023] Secondly, this invention also discloses a direction-aware 5D look-up table, which can use the geometric gradient information contained in the color difference channel separated in the previous step to guide the repair of the luminance channel. This design directly responds to the physical nature of the color channel spatial misalignment caused by LCA, and realizes directional, physically driven, precise correction, making the repair result sharper and cleaner.

[0024] Finally, this invention perfectly solves the problems of cost in hardware solutions and efficiency in general models. Its lightweight network design (only 0.13M parameters) can process 2K and even 4K ultra-high-definition images in real time on resource-constrained devices such as smartphones. Many large models will crash due to memory overflow in such scenarios, providing an unprecedented and truly feasible technical path for popularizing high-quality purple fringing removal functions on mobile devices.

[0025] The following section provides a detailed explanation of each module in the aforementioned device, using specific methods: Color perception transformation module: First, input an image. The image is fed into the Color Perceptual Transform (CA-CT) module. The core function of this module is to dynamically map the image from the standard RGB space to a customized color difference space (CAS). Specifically, it first receives a raw input image containing color differences. Internally, this module integrates a lightweight convolutional neural network (CNN) encoder. This encoder, based on the modern ConvNeXt architecture, is designed to efficiently extract global contextual information and latent color difference distribution patterns between pixels in the input image. Notably, this CNN encoder does not directly transform pixel values; instead, it dynamically predicts and generates a pixel value best suited to the current image based on its learned image features. The transformation matrix M is used for matrix multiplication. Through this linear transformation, the color values ​​of pixels are mapped to a completely new, three-dimensional color difference space (CAS), which can be mathematically expressed as follows: ; in This represents the three components in the CAS space. After the transformation, the color information of the original image is decoupled and redistributed to three mutually orthogonal channels: the luminance channel... (Primarily carries the image's brightness information and most structural details), color difference channel (Specifically designed to enrich chromatic aberration artifact information caused by LCA, separating it from normal image content), and positive traffic channels (It carries the remaining orthogonal color information and is usually preserved in subsequent processing). To ensure the physical stability and interpretability of this CAS space dynamically constructed by the learning algorithm, that is, to ensure that its three channels are as independent and orthogonal as possible in terms of information, this application introduces a key regularization term during training—axis alignment loss. The loss function aims to force the transformation of the three row vectors of the matrix (denoted as ). Geometrically, they tend to be orthogonal, meaning their dot product approaches zero. Their mathematical definition is as follows: Orientation perception correction module: Within the color difference space, perform core orientation-aware correction and brightness channel correction. The correction is accomplished by an innovative 5D Lookup Table (5D-LUT). Traditional 3D LUTs typically only index and look up based on RGB color values, while the 5D-LUT in this application is a more powerful color-space joint corrector with spatial awareness. For each pixel p in the luminance channel, its lookup coordinates in the 5D-LUT are determined by five dimensions, which integrate the pixel's own luminance information, its local spatial context, and the geometric characteristics of purple fringing artifacts. By introducing the geometric gradient information of the chrominance channel as an additional lookup dimension, the correction behavior of the 5D-LUT is no longer blind color mapping, but has gained precise direction awareness and spatial guidance capabilities. It can intelligently identify the specific direction and intensity of chrominance differences at the image edges and perform reverse, pixel-level precise compensation repair accordingly. This physically based design allows the 5D-LUT to completely eliminate purple fringing artifacts while preserving or even enhancing the original sharpness and structural details of the image edges to the greatest extent. At the same time, the chrominance channel... It itself uses a simpler learnable 1D lookup table (1D-LUT) for global mapping to suppress residual artifact intensity. To ensure that the learning process of the two lookup tables is smooth and stable, and to avoid introducing new mutations or artifacts, a smoothing loss is applied. It achieves regularization by penalizing the gradient between adjacent entries in the LUT.

[0026] Fusion and reverse mapping module: After calibration, the corrected luminance channel is obtained. ′ and color difference channel These two passages connect with the unaltered main traffic lanes. The process involves fusion, followed by the analytic inverse of the transformation matrix. Perform a reverse mapping to safely convert the image from the color difference space back to the RGB domain, obtaining the final output image. : ; The entire framework is optimized end-to-end through a composite objective function that integrates reconstruction fidelity and physical regularization terms: ; in, This is a comprehensive reconstruction loss used to ensure that the final output image is visually highly consistent with the ground truth image without chromatic aberration. It is designed as a combination of L1 pixel loss and a more advanced YCbCr perceptual loss. The YCbCr perceptual loss uses a pre-trained VGG19 network to extract deep features for comparison in the luminance (Y) channel to capture higher-level structural and semantic similarity, while directly using L1 distance for comparison in the chrominance (Cb, Cr) channels. This ensures accurate color control while maintaining the realism of image structure. Parameters and These are weighting coefficients used to balance the contributions of different loss terms to the total loss. Through this sophisticated loss function design and end-to-end training strategy, 5DLUT can stably and efficiently learn complex and accurate chromatic aberration removal mapping relationships.

[0027] Experimental test example: This invention underwent comprehensive and rigorous experimental evaluation on various test sets, including a specially constructed large-scale synthetic chromatic aberration dataset (PF-Synth) and a wide collection of real-world images. These experiments were compared across the board with several leading methods in the field of image inpainting, covering specialized models based on lookup tables (LUTs) (such as SRLUT and NILUT), high-performance general-purpose image inpainting models based on Transformer architectures (such as SwinIR and Restormer), and state-of-the-art generative methods based on diffusion models (such as DDRM and Lightendiffusion). The experimental results clearly demonstrate that the DCA-LUT of this invention achieves significant breakthroughs in chromatic aberration correction effectiveness, operational efficiency, and visual fidelity, reaching industry-leading levels.

[0028] Four widely accepted metrics are used to evaluate performance: PSNR, SSIM, LPIPS, and ΔE. PSNR and SSIM focus on evaluating the fidelity and structural integrity of the reconstructed image, while LPIPS evaluates perceptual quality by measuring the similarity of deep features, providing an assessment that is more in line with human visual quality.

[0029] Among them, PSNR (Peak Signal-to-Noise Ratio) is a commonly used metric for measuring image quality. It is mainly used to evaluate the difference between the processed image and the original image. The higher the value, the better the image quality. The calculation formula is: ; ; Where MSE represents mean squared error, which measures the difference between two images at the pixel level; I(i,j) and K(i,j) represent the pixel values ​​at position (i,j) of the generated fused image and the corresponding gold standard image (ground truth), respectively; m and n represent the width and height of the image, respectively; and MAX represents the maximum value of the image pixels.

[0030] SSIM (Structural Similarity Index Measure) is an indicator used to evaluate the quality of two images. It mainly measures the similarity of images by comparing their brightness, contrast, and structural information. The closer the value is to 1, the better the image quality. The formula for calculating SSIM is as follows: in, yes average value, Is with Corresponding gold standard The average value, It represents an image and covariance, yes variance yes The variance; L is the dynamic range of pixel values. and This represents the preset hyperparameters, here =0.01, =0.03, and This represents the smoothing parameter.

[0031] The LPIPS (Learned Perceptual Image Patch Similarity) metric is a measure used to evaluate the perceptual similarity between two images. It is based on features extracted by deep convolutional neural networks, rather than simple pixel differences; a lower LPIPS value indicates better image quality. The formula for calculating LPIPS is as follows: ; Where x represents the fused image output by the model, express The corresponding gold standard image, and This indicates that the first [unclear] extracted through a pre-trained VGG deep neural network... Layer feature map. Is the image x at the th Features of the layer Is the image x0 at the th... Characteristics of the layer. This represents the Euclidean distance (L2 norm) between feature maps. These are feature maps. and The difference metric between two images is used to measure the difference between them at the 1st minute. Similarity in layer features. This represents the weighting coefficients for the feature maps of each layer. Different network layers have different impacts on perceptual similarity, and the weights... This is typically reflected through training and learning. Indicates all layers The results are summed to comprehensively consider the differences in characteristics at different levels.

[0032] In addition to widely recognized industry-standard metrics, this application also innovatively proposes exclusive evaluation metrics specifically for the task of color difference correction.

[0033] ECAS (Edge Chromatic Aberration Score): General metrics such as PSNR and SSIM are often insensitive to local, color-specific artifacts like chromatic aberration, potentially resulting in high scores but still visible purple fringing. To address this, this application innovatively proposes ECAS, a task-specific metric, to accurately quantify residual chromatic aberration in high-contrast edge regions. The core idea of ​​ECAS is to first locate all high-contrast edges of interest using a high-quality gold-standard image, and then calculate the proportion of pixels in these edge regions that still exhibit purple or abnormal color shifts in the processed image. The calculation process consists of three steps: First, the gold-standard image is processed using a Sobel filter... Generate an edge mask in Secondly, the processed image Convert to HSV color space and generate a color difference mask based on preset hue and saturation thresholds. Finally, the ECAS score is calculated using the following formula; the lower the score, the cleaner the purple edge removal. Table 1 Quantitative comparisons on the PF-Synth dataset Quantitative comparative analysis: As shown in Table 1, the proposed method demonstrates industry-leading performance across multiple key image quality evaluation metrics, outperforming or matching the best current methods. On the SSIM metric, which measures image structure fidelity, the proposed method achieved the highest score of 0.9849, indicating that it effectively corrects chromatic aberration artifacts while preserving the structural and textural details of the original image most completely. Similarly, the proposed method also performs best on the LPIPS metric, which measures human visual perception quality, and the ΔE metric, which measures color reproduction accuracy, achieving the lowest perceptual difference of 0.0303 and the lowest chromatic aberration of 0.6784, respectively. This fully demonstrates that the images generated by the proposed method are visually the most natural and have the most realistic colors. Although the diffusion-based DDRM method achieved a slight lead in PSNR (39.154) and ECAS (0.0411), our method closely followed with a PSNR of 39.052 and an ECAS score of 0.0438, ranking second. This also confirms that our model possesses top-tier reconstruction fidelity and artifact removal capabilities. Particularly noteworthy is our ECAS score, a specific metric for purple fringing removal, which significantly outperforms powerful general-purpose models such as SwinIR (0.0755) and Restormer (0.0662). This strongly suggests that our specific design targeting the physical roots of the problem is more effective and accurate than general-purpose, large-scale models.

[0034] Model complexity and efficiency analysis: This application also comprehensively evaluated the computational complexity of the model. The results show that the method presented in this application achieves excellent performance while having a significant lead in efficiency. Regarding the number of model parameters, the method presented in this application has only 0.13M parameters, making it the second lightest model among all compared methods, far lower than models such as SwinIR (11.50M), Restormer (26.11M), and DDRM (72.52M), which often have tens of megabytes of parameters. In terms of inference speed, the method presented in this application performs best, processing 2K resolution images in only 0.0405 seconds and 4K ultra-high-definition images in only 0.1756 seconds, making it the fastest among all methods. Most importantly, thanks to its extreme efficiency and low memory consumption, the method presented in this application can easily process high-resolution images. In contrast, many powerful competitors, including SwinIR, Restormer, and DDRM, failed to process 2K or 4K images due to memory overflow, resulting in an "Out of Memory (OOM)" error and direct failure. This stark contrast demonstrates that the method presented in this application achieves the best possible trade-off between computational efficiency and restoration quality, and is the only practical solution that can provide a truly feasible and deployable solution for high-resolution image processing while ensuring top-notch results.

[0035] In one exemplary embodiment, the present invention also provides a computer device, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the image color difference correction method based on a 5D lookup table as described in Embodiment 1.

[0036] Those skilled in the art will understand that Figure 4 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0037] In one exemplary embodiment, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0038] In one exemplary embodiment, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0039] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0040] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0041] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units, etc., and are not limited to these.

[0042] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0043] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An image color difference correction method based on a 5D lookup table, characterized in that... Includes the following steps: Color perception transformation: Maps an image from RGB space to a color difference space. In the color difference space, image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. Direction-aware correction: In the luminance channel, a directional 5D lookup table is used to guide the luminance restoration by using the geometric gradient information extracted from the chrominance channel as part of the lookup index. The chrominance channel itself uses a 1D lookup table for global intensity suppression. Fusion and Inverse Mapping: The luminance and chrominance channel data, after dual-path correction, are re-fused with the unmodified orthogonal channel data. Then, through the inverse mapping operation of the transformation module, the image is converted from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

2. The image color difference correction method based on a 5D lookup table as described in claim 1, characterized in that: The original image is mapped to a color difference space by the color perception transformation module, and the artifacts of the image in the color difference space are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel.

3. The image color difference correction method based on a 5D lookup table as described in claim 2, characterized in that, The processing method of the color perception transformation module includes the following steps: First, receive a raw input image containing chromatic aberration. Within this module, a lightweight convolutional neural network (CNN) encoder is integrated. This CNN encoder dynamically predicts and generates an image that best fits the current image based on the image features it has learned. The transformation matrix M is used for matrix multiplication. Through linear transformation, the color value of the pixel is mapped to a color difference space CAS, which is expressed mathematically as follows: ； in This represents the three components in the CAS space; After the transformation, the color information of the original image is decoupled and redistributed to three mutually orthogonal channels: the brightness channel. Used to carry the brightness information and structural details of the image, as well as the color difference channel. It is used to enrich chromatic aberration artifact information caused by longitudinal chromatic aberration (LCA), separating it from normal image content, and positive channel. This is used to carry the remaining orthogonal color information and remains unchanged in subsequent processing.

4. The image color difference correction method based on a 5D lookup table as described in claim 3, characterized in that: A key regularization term, axis alignment loss, is introduced during the training of a convolutional neural network (CNN) encoder. This loss function is used to ensure that the three row vectors of the forced transformation matrix are geometrically orthogonal, and its mathematical definition is as follows: 。 5. The image color difference correction method based on a 5D lookup table as described in claim 1, characterized in that, The specific processing method for orientation perception correction includes the following steps: Within the color difference space, perform orientation-aware correction on the luminance channel. The correction is accomplished through a 5D lookup table. For each pixel p in the luminance channel, its lookup coordinates in the 5D lookup table are determined by five dimensions, which integrate the pixel's own luminance information, its local spatial context, and the geometric characteristics of purple fringing artifacts. Meanwhile, the chrominance channel... It uses a learnable 1D lookup table for global mapping to suppress residual artifact intensity.

6. The image color difference correction method based on a 5D lookup table as described in claim 5, characterized in that: To ensure that the learning process for 5D and 1D lookup tables is smooth and stable, a smoothing loss is applied. Regularization is achieved by penalizing the gradient between adjacent entries in the LUT: in, The 1D lookup table used to represent the color difference channels; This indicates the 5D lookup table used for the luminance channel; Denotes the square of the Frobenius norm; This represents a second-order difference operation on the weights of a 1D lookup table, used to smooth a one-dimensional curve. This represents the gradient operator that calculates the difference between adjacent weights along the d-th dimension, used to constrain the local continuity of the 5D space; summation symbol. This means calculating and summing the smoothness of each of the five dimensions of the 5D lookup table.

7. The image color difference correction method based on a 5D lookup table as described in claim 1, characterized in that, The fusion and reverse mapping include the following steps: After orientation-aware correction is completed, the corrected brightness channel is obtained. ′ and color difference channel These two passages connect with the unaltered main traffic lanes. The process involves fusion, followed by the analytic inverse of the transformation matrix. Perform a reverse mapping to convert the image from the color difference space back to the RGB domain, resulting in the final output image. : in, Color perception transformation matrix The inverse matrix; the right side of the matrix is ​​a column vector composed of the corrected components.

8. The image color difference correction method based on a 5D lookup table as described in claim 1, characterized in that, The method employs an end-to-end optimization through a composite objective function that integrates reconstruction fidelity and physical regularization terms: ； in, It is a comprehensive reconstruction loss used to ensure that the final output image is visually highly consistent with the color-aberration-free real image; parameters and These are weighting coefficients used to balance the contributions of different loss terms to the total loss; To smooth out the loss, This represents axis alignment loss.

9. An image color difference correction device based on a 5D lookup table, wherein the device employs the image color difference correction method based on a 5D lookup table as described in any one of claims 1-8, characterized in that... The device includes: Color perception transformation module: used to map the image from RGB space to a color difference space. In the color difference space, the image artifacts are decoupled and separated into an independent color difference channel. The main structure and brightness information of the image are retained in the brightness channel, and the orthogonal color information of the image is located in the orthogonal channel. Direction-aware correction module: In the luminance channel, it uses a direction-aware 5D lookup table and the geometric gradient information extracted from the chromatic difference channel as part of the lookup index to achieve directional guidance for luminance repair. The chromatic difference channel itself uses a 1D lookup table for global intensity suppression. The fusion and inverse mapping module is used to re-fuse the dual-corrected luminance and chrominance channel data with the unmodified orthogonal channel data. Through the inverse mapping operation of the transformation module, it converts the image from the chrominance space back to the standard RGB color space, ultimately producing and outputting a high-quality image. .

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the image color difference correction method based on a 5D lookup table as described in any one of claims 1-8.

Images