Image signal processing system and processing method

Abstract

The invention discloses an image signal processing system and method, and relates to the technical field of image signal processing, and the method comprises the steps: extracting structural features of a collected original image signal, and generating a region classification mask; predicting a processing path of each region according to the region classification mask; according to the processing path, performing corresponding compression operation on each area, and embedding corresponding structure identification information in compressed data; and during decoding, enhancement processing is performed on the structure key area according to the structure identification information, so that the image restoration quality is improved. The continuous adjustment of the channel reservation number and the quantization step size is realized, and the fidelity control capability of a complex structure region is enhanced. And the image restoration quality and the detail retention effect are improved, and a closed-loop optimization process between compression and enhancement is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of image signal processing, and particularly to an image signal processing system and a processing method. Background Art

[0002] In the field of image signal processing, compression and enhancement are two key processes for improving image storage efficiency and visual quality. Existing image compression methods mainly adopt a unified compression ratio or configure compression based on global statistical features to seek a balance between coding efficiency and visual quality. However, when dealing with regions with significantly different structures or complex details, such methods often cannot adjust the compression strategy specifically, and are prone to problems such as detail loss in key regions or insufficient compression in low-structure regions, affecting the quality of image reconstruction and the accuracy of subsequent processing.

[0003] Meanwhile, image enhancement technology exists as an independent process in most cases, and is usually repaired by edge filtering, super-resolution interpolation, or residual compensation after decoding. Due to the lack of structural semantic records in the compression stage, the enhancement module cannot obtain the structural information during the image compression process, and can only perform secondary estimation based on the decoded data, resulting in local jumps in the enhancement effect, excessive edges, or aggravated artifacts, and it is difficult to ensure the stability and consistency of the restoration quality.

[0004] In addition, existing methods mostly divide image regions based on fixed rules and use static strategies for channel control and quantization configuration, lacking pixel-level structural perception and adaptive adjustment capabilities. For regions with complex or significant structures, traditional compression methods cannot effectively allocate resources to ensure detail retention; for background regions, it is also difficult to dynamically increase the compression ratio to reduce redundant data. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problems solved by the present invention are: existing image compression and enhancement methods have problems such as weak structural perception ability, disconnection between compression and enhancement, easy distortion in key regions, and low compression efficiency in low-structure regions, and how to achieve pixel-level path prediction guided by structural features, dynamic regulation of compression parameters, and cross-stage collaborative transmission of structural information.

[0007] To solve the above technical problems, the present invention provides the following technical solution: An image signal processing system, comprising: An image acquisition module, including an image acquisition unit and a preprocessing unit, for acquiring an original image and completing noise suppression and normalization; A structure modeling module, including a feature extraction unit and a region division unit, the feature extraction unit is used to construct a structural feature vector, and the region division unit is used to generate a region classification mask; The path prediction module includes a partition analysis unit and a path allocation unit, which are used to extract the structural priority distribution features, neighborhood variability, and boundary correlation, and generate a path mapping matrix; The compression execution module includes a structure density calculation unit and a compression control unit. The structure density calculation unit is used to perform weighted fusion and normalization on the structural priority scores, structural complexity indicators, and boundary correlation indicators of each pixel, construct a structure density function, and output a structure density response value. The compression control unit is used to determine compression parameters based on the structure density response value and execute the compression strategy corresponding to the path. The compression parameters include an integer value corresponding to the number of channels to be retained and an integer value corresponding to the quantization step. The number of channels to be retained is obtained by multiplying the structure density function value by the preset maximum number of channels and rounding down. The quantization step is obtained by multiplying the complementary value of the structure density function value by the preset maximum quantization accuracy and rounding down. The complementary value is 1 minus the structure density function value; The structure embedding module includes an embedding construction unit and a data alignment unit. The embedding construction unit is used to generate a structure embedding vector, and the data alignment unit is used to perform pixel-level integration of the structure information, compressed data, and path information; The enhancement processing module includes an identification parsing unit and an enhancement execution unit. The identification parsing unit is used to extract the structure identification information and path type, and the enhancement execution unit is used to perform enhancement processing on the key structural regions.

[0008] As a preferred solution of the image signal processing method of the present invention, wherein: structural features are extracted from the acquired original image signal to generate a region classification mask; According to the region classification mask, the processing paths of each region are predicted; According to the processing path, corresponding compression operations are performed on each region, and the corresponding structure identification information is embedded in the compressed data; During decoding, according to the structure identification information, enhancement processing is performed on the key structural regions, thereby improving the image restoration quality.

[0009] As a preferred solution of the image signal processing method of the present invention, wherein: the extraction of structural features includes performing noise suppression and normalization on the acquired original image signal to obtain a preprocessed image; On the preprocessed image, multi-scale edge responses, multi-directional texture intensities, and frequency-domain residual significance heat maps are respectively extracted, and the edge responses, texture intensities, and significance heat maps are combined as the structural feature vector of the pixel; The generated region classification mask includes calculating the structural field similarity between pixels; the structural field similarity is calculated by using the cosine similarity algorithm for the structural feature vectors of any two pixels; pixels with feature response values in the top X percent of all pixels are selected as initial cores, and the feature response values are determined by the weighted average of the structural feature vectors, representing the structural response intensity or significance of the pixels. For each structural consistency region, calculate the regional significance mean and the structural feature variance, and assign a structural priority score to all pixels within the region. According to the structural priority scores, pixels are classified through an adaptive threshold to generate a region classification mask, and the pixels are divided into a structural key region, a general region, or a background region.

[0010] As a preferred embodiment of the method for processing an image signal according to the present invention, wherein: the process of predicting the processing paths of each region includes performing a structure-guided path assignment rule based on the distribution characteristics of the structural priorities, the neighborhood variability, and the boundary correlation. The distribution characteristics of the structural priorities are to extract the structural partition to which each pixel belongs, and statistically calculate the degree of dispersion between the structural priorities of all pixels within the structural partition and the mean of the structural priorities within the partition. The neighborhood variability represents calculating the change amount of the structural priorities of the pixels within the neighborhood relative to the structural priority of the current pixel within the neighborhood centered on the current pixel. The boundary correlation is for the boundary pixels in the structural key region, and calculating the difference in structural priorities between the boundary pixels and the pixels whose labels in the neighborhood do not belong to the structural key region.

[0011] As a preferred embodiment of the method for processing an image signal according to the present invention, wherein: the path assignment rule includes, when the pixel label in the region classification mask is the structural key region, extracting the structural partition to which the pixel belongs, statistically calculating the distribution dispersion of the structural priorities, calculating the variability of the structural priorities within the pixel neighborhood, and detecting its structural correlation with the adjacent regions. When the distribution dispersion reaches the structural complexity threshold, or the structural correlation reaches the boundary correlation threshold, an enhanced compression path is assigned. When the distribution dispersion is lower than the structural complexity threshold and the structural correlation is lower than the boundary correlation threshold, a denoising and enhanced compression path is assigned. When the pixel label is the general region, an enhanced compression path or a compression path is respectively assigned according to whether the neighborhood structural priority variability reaches the detail enhancement threshold. When the pixel label is the background region, a denoised compression path is assigned.

[0012] As a preferred solution of the image signal processing method of the present invention, wherein: the compression operation includes constructing a structure density function according to the output result of the path allocation rule, in combination with the structure priority score, structure complexity index and boundary correlation index of each pixel; Determine the compression parameter through the structure density response value output by the structure density function; The compression parameters include an integer value corresponding to the number of channels to be retained and an integer value corresponding to the quantization step size.

[0013] The number of channels to be retained is obtained by multiplying the structure density function value by the preset maximum number of channels and taking the integer part after discarding the decimal part. The quantization step size is obtained by multiplying the complementary value of the structure density function value by the preset maximum quantization precision and taking the integer part after discarding the decimal part, where the complementary value is 1 minus the structure density function value.

[0014] As a preferred solution of the image signal processing method of the present invention, wherein: during the compression process, structure identification information is synchronously constructed, and the structure identification information includes the structure density response value, the compression parameter corresponding to the selected path type, and the compression residual information; Align and integrate the structure identification information in the form of an embedded vector with the compressed data and the path mapping matrix according to the pixel index to form a compressed output set.

[0015] A computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the image signal processing method.

[0016] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the image signal processing method are implemented.

[0017] Advantages of the present invention: The image signal processing method provided by the present invention can achieve fine division of the key structure areas, general areas and background areas in the image by introducing structure feature modeling and region classification masks, improving the processing granularity and perception accuracy. Combining the structure priority distribution, neighborhood variability and boundary correlation, a path prediction mechanism is constructed to realize the dynamic scheduling of the processing path. Constructing a structure density function as the core of compression parameter control, realizing continuous adjustment of the number of channels to be retained and the quantization step size, and enhancing the fidelity control ability for complex structure areas. During the compression process, a structure embedding vector is generated synchronously, integrating the structure response, path type and residual information into the compressed data to realize end-to-end transmission of structure information. In the decoding stage, targeted enhancement processing can be performed according to the structure identification information, performing residual compensation or structure-sensitive restoration in the key structure areas, thereby improving the image restoration quality and detail retention effect, and constructing a closed-loop optimization process between compression and enhancement. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0019] Figure 1 It is the overall flowchart of a method for processing an image signal provided in the first embodiment of the present invention. Specific Embodiments

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings of the specification. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0021] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for processing an image signal, including: S1: Extract structural features from the acquired original image signal to generate a region classification mask.

[0022] Taking the acquired original image signal as input, first perform noise suppression and normalization on the image to obtain a preprocessed image . Calculate the multi-scale edge response , the multi-directional texture intensity , and the frequency-domain residual significance heat map respectively on . For each pixel, combine , , as a feature vector : ; where x represents the coordinate of the pixel in the horizontal direction; y represents the coordinate of the pixel in the vertical direction; represents the two-dimensional spatial position of the current pixel in the image; represents the horizontal and vertical coordinates of the first pixel participating in the structural similarity calculation; represents the horizontal and vertical coordinates of the second pixel participating in the structural similarity calculation.

[0023] For all pixel feature vectors, calculate the structural field similarity one by one: ; Among them, represents the structural feature similarity between position and position . represents the structural feature vector at position . represents the structural feature vector at position . represents transpose.

[0024] Pixels with feature response values in the top X percent (e.g., the top 5%) of all pixels are selected as the initial cores, and density adaptive clustering is performed based on the similarity threshold to successively expand and obtain the structure consistency region . Each pixel belongs to only one clustering region. For each clustering region , calculate the significance mean and the structural feature variance of this region, and assign a structural priority score to all pixels in the region. The formula is expressed as: ; Among them, represents the structural priority score of pixel , represents the significance mean of region , represents the variance of the structural features within region , represents the maximum value of the structural feature variances in all regions, represents the pixel coordinates, represents the th structural partition.

[0025] Based on , by setting the thresholds , , each pixel is divided into a structural key area, a general area, or a background area to obtain the region classification mask : ; Among them, represents the region classification label of pixel point , taking one of the values 2, 1, 0; represents the adaptive segmentation threshold between the structural key area and the general area; represents the adaptive segmentation threshold between the general area and the background area; 2 represents the structural key region, 1 represents the general region, and 0 represents the background region.

[0026] Finally, a region classification mask is obtained. The size of this mask matrix is the same as that of the input image, providing a basis for subsequent adaptive path prediction and compression processing based on the structure field.

[0027] During the process of structural feature extraction and region classification mask generation, by implementing noise suppression and normalization on the original image signal, environmental interference and signal deviation during the acquisition process can be effectively excluded, providing a unified basis for subsequent structural feature segmentation. Multi-scale edge response, texture intensity, and frequency-domain saliency features are jointly extracted in the same image space, significantly enhancing the resolution ability of local contours, texture patterns, and salient regions. The vectorized description method based on structural field features realizes the comprehensive quantification of complex structural distributions, laying a foundation for subsequent spatial partitioning and structure recognition.

[0028] Furthermore, in the stage of structural similarity calculation and density adaptive clustering, the cosine similarity of structural feature vectors is used to measure the spatial structure relationship between pixels, replacing the traditional methods of linear weighted fusion and simple threshold segmentation in the past. This mechanism supports the dynamic coupling expression between high-dimensional structural features. Through iterative clustering, pixel sets with highly consistent structures are automatically merged into adaptive regions, achieving the precise segmentation and deformable partitioning of the local structure continuity of the image. This process overcomes the problems of insufficient generalization ability and poor robustness of conventional partitioning methods based on single features or fixed templates when dealing with multi-scale and multi-type structures.

[0029] Even further, in the hierarchical mask generation link, through the comprehensive scoring of structural significance mean and clustering variance, the importance and complexity of regions can be adaptively sorted. The finally output region classification mask can not only accurately distinguish key structural regions, general regions, and background regions, but also dynamically adapt to image inputs of different types and different structural complexities. The overall scheme breaks through the barrier that traditional image partitioning can only handle regular regions and is difficult to adapt to structural diversity, realizing an all-round innovation from pixel-level feature expression, spatial structure clustering to multi-level differentiation, effectively improving the accuracy and robustness of subsequent path prediction and structure adaptive compression processing.

[0030] S2: According to the region classification mask, predict the processing paths of each region.

[0031] First, for each pixel , distinguish the region type it belongs to according to the mask label . For the key structural region, further analyze the degree of local structural change of the key structural region. In each neighborhood centered on , calculate the local variability of the structural priority: ; Among them, represents the local structure priority variability of the pixel ; N represents the set of neighboring pixels centered on the pixel ; represents the structure priority score of the pixel within the neighborhood; represents the structure priority score of the current pixel ;

[0032] In the structure key area, for each pixel, extract its belonging clustering partition , and calculate the structure complexity index based on the dispersion degree of the structure priority distribution within the same partition: ; Among them, represents the structure complexity score of the pixel located at the position ; represents the structure clustering area to which the pixel belongs, with the index ; represents the total number of pixel points (i.e., the cardinality of the set) in the structure area ; represents any pixel position index in the clustering area ; represents the structure priority score of the pixel ; represents the mean value of the structure priorities of all pixels within the structure area .

[0033] For the pixels at the boundary of the structure key area, that is, there are pixels with labels not equal to 2 in their neighborhoods, a structure correlation index is further introduced. This index measures the structural feature difference between the boundary pixels and the adjacent areas, and the specific definition is as follows: ; ; Among them, represents the set of all pixel points centered on the pixel with labels not equal to 2 in the neighborhood; represents the abscissa and ordinate of any pixel point in the neighborhood; represents the neighborhood set centered on the pixel ; represents the area classification label of the pixel point ; represents that this pixel point does not belong to the structure key area; represents the pixel Structural correlation index; Denotes taking the maximum value of all items in the set; Denotes selecting pixels from the set of pixels in the non-structural key area of the neighborhood.

[0034] Pixels with high boundary correlation are more likely to have cross-region structural breaks or detail mismatches. Higher enhancement and collaborative processing priorities should be assigned during path allocation.

[0035] During the process of processing path allocation, define the structural complexity threshold , which determines the enhancement requirements for complex structural areas; the boundary correlation threshold , which distinguishes high-difference boundaries from ordinary boundaries; the general area priority variability threshold , which is used for detail compensation.

[0036] These thresholds can be set through partition statistics, image scene experience, or adaptive optimization.

[0037] For each pixel, , combining the above structural indicators and region labels, allocate the signal processing path according to the following rules: If it satisfies, , and at the same time satisfies, , or, , it indicates that the pixel is located in a complex structure or boundary mutation area. First enhance its details and then perform compression. The allocated path is "compress after enhancement".

[0038] If it satisfies, , and at the same time satisfies, , and, , it indicates that the structure distribution is stable. To prevent weak structure textures from being distorted during compression, first perform denoising, then perform enhancement and compression. The allocated path is "denoise, enhance, and then compress".

[0039] If it satisfies, , and, , it indicates that the details in the general area are complex. The fidelity before compression should be improved through enhancement. The allocated path is "compress after enhancement".

[0040] If it satisfies, , and, , the structural change is limited. Directly perform the compression operation. The allocated path is "compress".

[0041] If it satisfies, , that is, the pixel is located in the background area, which usually does not contain significant structural information. To improve the compression efficiency, first perform denoising to reduce redundancy, and then perform the compression operation. The allocated path is "compress after denoising".

[0042] The processing path of each pixel is written into the path allocation matrix in the form of a task identification sequence, , and the matrix size is consistent with the original image, serving as the scheduling basis for subsequent compression and enhancement processes, achieving fine control and dynamic allocation of paths based on structural features.

[0043] By constructing two types of structural behavior indicators, namely structural complexity and boundary correlation, the image structure state can be further refined within the same region label, avoiding the problem that traditional region classification methods lack recognition and response to internal structural features after label division. The combined discrimination method of local variability, clustering distribution dispersion, and cross-region priority difference enhances the sensitivity of the path prediction process to complex textures, edge attenuation, and multi-scale structures, providing a higher-resolution decision basis for compression and enhancement strategy allocation.

[0044] Furthermore, during the path allocation process, by introducing a five-level task process guided by structure, the signal processing path can be flexibly switched according to the functional roles of different regions in the image structure field. Compared with traditional strategies that only support static processing sequences such as "compression" or "compression + enhancement", this mechanism supports various combinations such as "denoising → enhancement → compression" and "enhancement → compression", and can dynamically adjust the path level within the region. The path matrix construction method driven by structural judgment results realizes the binding of pixel-level processing flow scheduling and execution units, providing a logical basis for subsequent adaptive compression.

[0045] Even further, when dealing with the intersection problem between the boundary region and the general region, by using the correlation analysis of the maximum priority difference, the problem of reconstruction artifacts caused by path breaks and structural discontinuities between regions in existing methods is effectively solved. The collaborative enhancement strategy driven by cross-region structural differences enables complex regions and adjacent regions to complete detail consistency compensation before compression, breaking through the limitations of traditional path allocation strategies at the region boundary, such as fixed strategies, lagging responses, and rigid processing, and realizing the compression and enhancement collaborative scheduling ability for structurally continuous regions.

[0046] S3: According to the processing path, perform corresponding compression operations on each region, and embed the corresponding structural identification information in the compressed data.

[0047] Combined with the structural priority score , the structural complexity index and the boundary correlation index , construct a structure-driven compression mapping function, and execute different compression strategies according to different path types. First, to dynamically respond to the compression fidelity requirements of structural regions, introduce the structural density function to control the channel retention rate and quantization accuracy: ; Among them, Represents pixels The structural response density at ; Represents the Sigmoid function; , , Respectively represent the structural priority , Structural complexity Correlation with boundaries Weight coefficient for structural density; Represents the structural priority score of the pixel; Represents the structural complexity index within the partition to which it belongs; Indicates the structural boundary correlation between the point and the adjacent area.

[0048] In the actual compression process, σ calculates the number of compression channels and the quantization step size according to μ(x, y): ; in, Represents pixels The corresponding number of compression channels retained; Indicates the corresponding quantization step size; Indicates the maximum number of compression channels that can be allocated; Indicates the maximum configurable quantization accuracy; Indicates a round-down operation. This mechanism allows areas with high structural density to retain more feature channels and reduce quantization errors, while areas with low structural density automatically shrink the coding capacity to improve compression efficiency.

[0049] The compression module adopts a split execution strategy for different path types. If the path is "enhancement followed by compression" or "denoising, enhancement followed by compression", it is preferred to retain more structural feature dimensions before compression and use a lower quantization step size. If the path is "compression" or "denoising followed by compression", the compression ratio is increased and the number of retained channels is reduced to improve the overall compression rate.

[0050] The structure embedding vector is constructed synchronously during compression: ; in, Represents the output value of the structure density function, represents the channel compression residual, represents the quantization error response.

[0051] The embedding vector is written in full form in the key area of ​​the structure and in compressed form in the general area. The embedding field can be omitted in the background area to reduce data redundancy.

[0052] The final output consists of compressed encoded data, structure embedded information And the path mapping matrix It consists of, and each piece of data is aligned and integrated according to the pixel position during the compression process. To uniformly represent the output content, the following structure compression output set is defined: ; Among them, represents the encoded result after compression, s is the structure embedding vector, is the path indication information.

[0053] During the compression process, a structure density function is introduced, effectively solving the problem that traditional image compression methods cannot adjust compression parameters according to the structural differences in image regions. Traditional methods often use a unified compression ratio or rule-based compression based on static region division, resulting in over-compression in key structural regions and under-compression in low-structure regions, affecting the overall image restoration quality. By constructing a structure density function with structure priority score, structure complexity, and boundary correlation as inputs, the structure response value at each pixel position can be continuously calculated, and accordingly, the number of channels retained and the quantization step size can be dynamically adjusted, enabling the compression process to finely adapt to the image structure characteristics and realizing the linkage adjustment between the compression intensity and the image structure complexity.

[0054] On the basis of controlling the compression parameters, a structure embedding vector mechanism is introduced to avoid the problem that structural information is completely discarded during the compression process. Traditional compression methods usually only retain pixel coding information, and the structural semantics need to be re-estimated at the decoding end, resulting in a lack of path and structure prior support in the enhancement stage. The structure embedding vector consists of three parts: the structure density function value, the channel compression residual, and the quantization error. It can record the response relationship between the compression behavior and the structural characteristics, and is encoded integrally with the compressed data, directly providing a structure enhancement reference on the decoding side and improving the pertinence and consistency of structure restoration.

[0055] The compressed output data adopts a combined form of pixel alignment of the path mapping matrix, the compression result, and the structure embedding vector, effectively breaking through the problem of the separation of the traditional compression-enhancement processing chain. In the original scheme, the compression and enhancement processes are independent of each other, lacking a cross-stage structural semantic transmission mechanism, and it is difficult to construct a structure field-driven enhancement logic. By unifying the compressed output structure, the joint output of the path, structure density, and compression behavior is realized, providing continuous structural semantic input for the subsequent enhancement module, enabling the entire image processing chain to have the ability of structure adaptive processing, and making the enhancement restoration effect more delicate and stable.

[0056] S4: During decoding, according to the structure identification information, perform enhancement processing on the key structural regions of the structure, thereby improving the image restoration quality.

[0057] Extract the structure embedding vector at each pixel position , the vector contains the structure density response values, channel compression residuals, and quantization error information recorded during the compression process. Based on the structure density values , estimate the structural fidelity of the pixels during the compression stage, which serves as the initial basis for determining whether the current pixels enter the enhancement processing flow.

[0058] For the pixels in the structure key area that satisfy , further combine the structure density response value with the path label to determine the intensity level of the enhancement strategy. If the path type is "compressed after enhancement" or "denoised, compressed after enhancement", it indicates that the original structure has been strengthened before compression. At this time, during decoding enhancement, residual recovery should be preferentially performed using the compression residual information and the quantization error information as reference values to correct the output of the main decoding channel; if the path type is "compressed" or "compressed after denoising", it indicates that this area has not undergone strong enhancement processing, then enable the structure-sensitive filter to reconstruct local details, and use as the guiding factor to dynamically adjust the response range of the enhancement kernel.

[0059] The enhancement process uses a pixel-level adaptive filtering module. For each pixel in the structure key area, select the enhancement method according to the following logic: When the structure density value is higher than the set enhancement start threshold , perform structure response-guided saliency enhancement.

[0060] When the structure density value is in the middle range, compensate for the channel decoding error by combining the compression residual information.

[0061] When the structure density value is lower than the set threshold, skip the enhancement process and only perform conventional interpolation restoration.

[0062] The enhanced pixel values are weighted and fused with the original decoded pixel values. The fusion weight is controlled by the output value of the structure density function to achieve the joint update of compression residual repair and structure-guided enhancement. The enhancement result is kept consistent with the original image size, and the enhanced image frame is output pixel by pixel.

[0063] The final enhanced output is dominated by the structure key area, reconstructing texture, contour, and edge features in the significant area, effectively offsetting the loss of high-frequency information during the compression process, and improving the structural fidelity and detail quality of the overall restored image.

[0064] Embodiment 2 is an embodiment of the present invention, which provides a method for processing an image signal. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0065] Select 60 high-resolution images (1024×768) from the COST2100 visual scene database, including people, buildings, and natural environments, as the evaluation sample set. The experimental platform uses an NVIDIA RTX 3080 GPU + Intel i9-12900KF processor, and Python + CUDA is used to accelerate the algorithm. The original images are first processed by the image acquisition module to complete YUV channel separation and brightness normalization. After unifying the image size, they are sent to the structure modeling module. The structure feature extraction unit calculates the multi-scale gradient direction distribution, texture coherence, and frequency domain significance response of the images to generate structure feature vectors; the region division unit uses the density clustering method (DBSCAN) combined with the structure field similarity metric to generate a structure classification mask, separating the key structure regions, general regions, and background regions.

[0066] The path prediction module constructs a three-dimensional feature tensor using the structure priority score variance, neighborhood deviation coefficient, and edge jump amplitude, and combines it with a heuristic partition scoring function to calculate the path mapping per pixel. The compression execution module adjusts the channel allocation and quantization bit width through the structure density function, controls the compression intensity per pixel, and forms a dynamic compression mapping diagram. During the compression process, a structure embedding vector is generated synchronously and integrated into the output bitstream. After decoding, the enhancement module performs residual inversion and significance-aware enhancement operations according to the path type and structure density response value. The entire process performs batch processing on 60 images, and the average processing time per image is 1.26 seconds.

[0067] Average size of the original images: 746.21 KB.

[0068] Traditional mean block compression method (fixed path): PSNR: 27.84 dB, SSIM: 0.8013, average PSNR of the key structure regions: 25.66 dB, file size: 132.15 KB.

[0069] WebP image compression algorithm (Google): PSNR: 29.92 dB, SSIM: 0.8489, average PSNR of the key structure regions: 27.11 dB, file size: 124.50 KB.

[0070] The described structure-guided system: PSNR: 31.34 dB, SSIM: 0.8796, average PSNR of the key structure regions: 29.76 dB, file size: 120.03 KB.

[0071] Average improvement amplitude after processing by the image enhancement module (key structure regions): PSNR is increased by 2.65 dB, and the edge restoration accuracy is increased by 14.32%.

[0072] The compression rate is increased by 3.59% compared to the WebP algorithm, and the detail recognition score after enhancement is increased by 18.45%.

[0073] The experimental results show that the structure field-guided compression enhancement system is superior to the existing mainstream solutions in terms of compression ratio, overall image quality, and the restoration of details in the structural area. The traditional method with a fixed compression path performs poorly in restoring details in key areas, suffers serious loss of edge information, and the texture in the structural jump area is prone to blurring. Although the WebP algorithm adopts lossy compression optimization, it does not model the structural features of the image, and is prone to cause detail fractures in the high-frequency structural area.

[0074] The system identifies regions of the image display structure through the structure modeling module, and combines the path prediction mechanism to allocate the optimal processing chain for different regions, avoiding the structural confusion problem caused by "uniform compression". The adjustment of compression parameters under the control of the structure density function makes the number of channels in the structurally dense area increase and the quantization step size converge, significantly improving the structural fidelity. The structure embedding mechanism encapsulates structural semantics and compression behavior synchronously during the compression stage, and the enhancement module can accurately perform targeted repair, avoiding the miscompensation caused by the distorted estimation of the structural position in traditional enhancement algorithms.

[0075] Especially in fine structural areas such as building edges and human hair, the enhancement module guides the filtering range and intensity based on the structural density response, realizes hierarchical restoration of details, and effectively solves the problems of edge distortion and ghosting caused by compression. While ensuring an increase in the compression rate, the PSNR of the structural area of the overall system is increased by 2.65 dB compared with WebP, and the subjective detail score is increased by nearly 18%, verifying the effectiveness and technological foresight of the structure-driven mechanism in the collaborative optimization of image compression and enhancement.

[0076] Embodiment 3, an embodiment of the present invention, provides an image signal processing system, including: An image acquisition module, including an image acquisition unit and a preprocessing unit, for acquiring an original image and completing noise suppression and normalization.

[0077] A structure modeling module, including a feature extraction unit and a region division unit, the feature extraction unit is used to construct a structural feature vector, and the region division unit is used to generate a region classification mask.

[0078] A path prediction module, including a partition analysis unit and a path allocation unit, for extracting structural priority distribution features, neighborhood variability, and boundary correlation, and generating a path mapping matrix.

[0079] A compression execution module, including a structure density calculation unit and a compression control unit, the structure density calculation unit is used to construct a structure density function and output a structure density response value, and the compression control unit is used to calculate compression parameters and execute the compression strategy corresponding to the path.

[0080] The structure embedding module includes an embedding construction unit and a data alignment unit. The embedding construction unit is used to generate a structure embedding vector, and the data alignment unit is used to perform pixel-level integration of the structure information, the compressed data, and the path information.

[0081] The enhancement processing module includes an identification parsing unit and an enhancement execution unit. The identification parsing unit is used to extract the structure identification information and the path type, and the enhancement execution unit is used to perform enhancement processing on the key areas of the structure.

[0082] If a function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs and other various media that can store program codes.

[0083] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0084] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which a program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or, if necessary, other suitable processing, and then storing it in a computer memory.

[0085] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, the multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An image signal processing system, characterized in that, Including: An image acquisition module, including an image acquisition unit and a preprocessing unit, for acquiring an original image and completing noise suppression and normalization; A structure modeling module, including a feature extraction unit and a region division unit, where the feature extraction unit is used to construct a structure feature vector, and the region division unit is used to generate a region classification mask; A path prediction module, including a partition analysis unit and a path allocation unit, for extracting structure priority distribution features, neighborhood variability, and boundary correlation, and generating a path mapping matrix; A compression execution module, including a structure density calculation unit and a compression control unit. The structure density calculation unit is used to perform weighted fusion and normalization on the structure priority score, structure complexity index, and boundary correlation index of each pixel, construct a structure density function, and output a structure density response value. The compression control unit is used to determine compression parameters based on the structure density response value and execute the compression strategy corresponding to the path, where the compression parameters include an integer value corresponding to the number of channels to be retained and an integer value corresponding to the quantization step. The number of channels to be retained is obtained by multiplying the structure density function value by the preset maximum number of channels and rounding down. The quantization step is obtained by multiplying the complementary value of the structure density function value by the preset maximum quantization accuracy and rounding down, where the complementary value is 1 minus the structure density function value; A structure embedding module, including an embedding construction unit and a data alignment unit. The embedding construction unit is used to generate a structure embedding vector, and the data alignment unit is used to perform pixel-level integration of structure information, compressed data, and path information; An enhancement processing module, including an identification parsing unit and an enhancement execution unit. The identification parsing unit is used to extract structure identification information and path types, and the enhancement execution unit is used to perform enhancement processing on key structure regions; 2. The processing method of an image signal processing system according to claim 1, characterized in that: Extract structure features from the acquired original image signal to generate a region classification mask; Predict the processing paths of each region according to the region classification mask; Perform corresponding compression operations on each region according to the processing path, and embed corresponding structure identification information in the compressed data; During decoding, perform enhancement processing on key structure regions according to the structure identification information, so as to improve the image restoration quality; 3. The method for processing an image signal according to claim 2, wherein: The extraction of structure features includes performing noise suppression and normalization on the acquired original image signal to obtain a preprocessed image; Extract multi-scale edge responses, multi-directional texture intensities, and frequency domain residual significance heat maps on the preprocessed image respectively, and combine the edge responses, texture intensities, and significance heat maps as the structure feature vector of the pixel; The generated region classification mask includes calculating the structural field similarity between pixels based on the structural feature vector; the structural field similarity is calculated by using the cosine similarity algorithm for the structural feature vectors of any two pixels; pixels with feature response values in the top X percent of all pixels are selected as initial cores, and the feature response value is determined by the weighted average of the structural feature vectors, representing the structural response intensity or significance of the pixel. For each structural consistency region, calculate the regional significance mean and the structural feature variance, and assign a structural priority score to all pixels within the region. According to the structural priority score, classify the pixels through an adaptive threshold to generate a region classification mask, and divide the pixels into a structural key region, a general region, or a background region.

4. The method for processing an image signal according to claim 3, wherein: The process of predicting the processing path for each region includes executing a structure-guided path allocation rule based on the distribution characteristics of the structural priority, the neighborhood variability, and the boundary correlation. The distribution characteristic of the structural priority is to extract the structural partition to which each pixel belongs, and statistically calculate the degree of dispersion between the structural priorities of all pixels within the structural partition and the mean of the structural priorities within the partition. The neighborhood variability represents calculating the change amount of the structural priority of the pixels within the neighborhood relative to the structural priority of the current pixel within the neighborhood centered on the current pixel. The boundary correlation is for the boundary pixels in the structural key region, calculating the difference in structural priority between the boundary pixel and the pixels within the neighborhood whose labels do not belong to the structural key region.

5. The method for processing an image signal according to claim 4, wherein: The path allocation rule includes, when the pixel label in the region classification mask is the structural key region, extracting the structural partition to which the pixel belongs, statistically calculating the distribution dispersion of the structural priority, calculating the variability of the structural priority within the pixel neighborhood, and detecting its structural correlation with the adjacent region. When the distribution dispersion reaches the structural complexity threshold, or the structural correlation reaches the boundary correlation threshold, assign an enhanced compression path. When the distribution dispersion is lower than the structural complexity threshold and the structural correlation is lower than the boundary correlation threshold, assign a denoising, enhanced compression path. When the pixel label is the general region, assign an enhanced compression path or a compression path respectively according to whether the neighborhood structural priority variability reaches the detail enhancement threshold. When the pixel label is the background region, assign a denoising after compression path.

6. The method for processing an image signal according to claim 5, characterized in that: The compression operation includes constructing a structural density function based on the output result of the path allocation rule, combining the structural priority score, the structural complexity index, and the boundary correlation index of each pixel. Determine the compression parameter through the structural density response value output by the structural density function. The compression parameter includes an integer value corresponding to the number of channels to be retained and an integer value corresponding to the quantization step. The number of channels to be retained is obtained by multiplying the structural density function value by the preset maximum number of channels and discarding the decimal part to take the integer, and the quantization step is obtained by multiplying the complementary value of the structural density function value by the preset maximum quantization accuracy and discarding the decimal part to take the integer, where the complementary value is 1 minus the structural density function value.

7. The method for processing an image signal according to claim 6, wherein: Construct the structure identification information synchronously during the compression process, where the structure identification information includes a structure density response value, compression parameters corresponding to the selected path type, and compression residual information; Integrate the structure identification information in the form of an embedded vector, the compressed data, and the path mapping matrix in alignment with the pixel index to form a compressed output set.

8. The method for processing an image signal according to claim 7, wherein: The enhancement processing includes extracting the structure identification information and the path mapping matrix during the decoding process, and determining whether to perform enhancement processing by combining the region label corresponding to the pixel; When the path type is post-enhancement compression or denoising, post-enhancement compression, perform error compensation-based enhancement processing based on the compression residual information and quantization error information in the structure identification information; When the path type is compression or post-denoising compression, perform structure sensitivity enhancement processing based on the structure density information in the structure identification information, and adjust the application range of the enhancement processing according to the structure density response value; Fuse the enhancement result and the decoded image at the pixel level, where the fusion weight is determined by the structure density response value, and output the enhanced image frame.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for processing an image signal according to any one of claims 2 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for processing an image signal according to any one of claims 2 to 8.

Images