Image effect enhancement method and system

Abstract

The invention belongs to the technical field of image processing, particularly relates to an image effect enhancement method and system, and aims to improve the image quality through multi-domain collaborative adaptive processing. The method comprises the following steps: firstly, determining an optimal quantization level through content-aware adaptive quantization, generating a quantization image and a region complexity graph, then decomposing the image into smooth, edge and texture sub-bands through multi-scale domain analysis, constructing a characteristic spectrum, then, applying histogram equalization, anisotropic diffusion and nonlinear sharpening to different sub-bands through adaptive characteristic enhancement, and finally, carrying out adaptive characteristic enhancement on the different sub-bands. And residual error optimization is carried out. And finally, multi-domain collaborative fusion is carried out to construct an enhancement matrix, fusion masks and feature correction are applied to obtain a final enhanced image, and the method is suitable for various images, retains details, enhances different regions in a targeted manner, and has a remarkable effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to an image effect enhancement method and system. Background Art

[0002] With the widespread use of digital imaging, the demand for image quality enhancement technologies is growing. In various application scenarios, such as medical imaging, video surveillance, mobile device imaging, and satellite remote sensing, raw images often suffer from poor visual quality due to insufficient lighting, noise interference, and blurred details. Therefore, efficient and reliable image enhancement technologies are urgently needed.

[0003] Existing image enhancement technologies are primarily categorized into spatial and frequency domain methods. Spatial domain methods, such as histogram equalization and filter enhancement, directly process pixels; frequency domain methods, such as Fourier transform and wavelet transform, perform enhancement in the transform domain. These methods each have their advantages, but also share common limitations.

[0004] For example, Chinese patent CN112233037B discloses an image enhancement system and method based on image segmentation. This method converts the image into a signal after ternary processing for enhancement. Although this method improves efficiency compared to traditional pixel processing methods, it still has the following shortcomings:

[0005] 1. A large amount of image detail information is inevitably lost during the ternary process;

[0006] 2. The preset fixed pixel interval lacks adaptability and is difficult to adapt to different types of images;

[0007] 3. The mathematical functions used lack adaptability to different image contents;

[0008] 4. The amplitude amplification parameter range is fixed and difficult to flexibly adjust according to image characteristics;

[0009] 5. Over-reliance on single-domain processing, failing to fully combine the respective advantages of spatial and frequency domains;

[0010] 6. There are many conversion steps in the overall processing flow, and the computational complexity is high.

[0011] These problems limit the application effect of existing technologies in complex scenes, especially in terms of detail preservation, noise suppression and visual naturalness. Summary of the Invention

[0012] The purpose of the present invention is to provide a multi-domain collaborative adaptive image enhancement method and system, aiming to overcome the shortcomings of the existing technology and achieve higher quality and more efficient image enhancement effects by integrating multiple domain processing technologies and introducing adaptive mechanisms.

[0013] The present invention proposes a multi-domain collaborative adaptive image enhancement method, comprising:

[0014] Get the image to be processed;

[0015] Perform content-aware adaptive quantization to obtain a quantized image and a regional complexity map, including: determining the optimal quantization level based on the local entropy and global contrast of the image; and determining quantization parameters for different regions based on the regional complexity index;

[0016] Perform multi-scale domain analysis and decomposition to obtain a multi-scale and multi-directional sub-band set, including: decomposing the quantized image into smooth sub-bands, edge sub-bands, and texture sub-bands; extracting the main eigenvectors of each sub-band and constructing a sub-band feature map;

[0017] Perform adaptive feature enhancement and reconstruction to obtain a set of enhanced subbands, including: applying adaptive histogram equalization to smooth subbands; applying anisotropic diffusion to edge subbands; applying nonlinear sharpening to texture subbands; and performing multi-scale prediction residual optimization.

[0018] Perform perceptual quality assessment and feedback optimization to obtain optimization parameters and regional importance weight maps, including: calculating the no-reference image quality assessment index; dynamically adjusting enhancement parameters based on the assessment results;

[0019] Perform multi-domain collaborative fusion to obtain the final enhanced image, including: constructing inter-domain complementary enhancement matrices; applying edge-aware fusion masks; and performing feature consistency correction.

[0020] Preferably, the content-aware adaptive quantization specifically includes: obtaining the local entropy and global contrast of the image; determining the quantization level n based on the local entropy and global contrast of the image, where 4≤n≦8; and calculating the regional complexity index. Among them, EdgeDensity is the edge density, which indicates the proportion of edge pixels in the image area; GradVariance is the gradient variance, which indicates the degree of change of pixel gradients in the area; MeanIntensity is the average brightness, which indicates the average value of pixel brightness in the area; is the resolution correction factor, expressed as , W and H are the width and height of the current image, and is the reference standard resolution; based on the regional complexity index, an adaptive weight w(x,y) is generated; and a quantization operation is performed: ,Get quantized images and regional complexity maps.

[0021] Preferably, the perceptual quality assessment and feedback optimization specifically include: calculating the non-reference image quality evaluation index NIQE = f (contrast, clarity, naturalness, artifact suppression); constructing a regional importance weight map ; Based on the evaluation index and weight map, construct the evaluation-parameter mapping relationship Iterative optimization of processing parameters via stochastic gradient descent ,in, P is the comprehensive cost function, NIQE(P) is the no-reference quality evaluation score (normalized to between 0 and 1), Artifact(P) is the artifact evaluation function, Detail(P) is the detail preservation evaluation function, and α, β, and γ are the weight coefficients for balancing various indicators. Iterative optimization continues until or the maximum number of iterations is reached.

[0022] As an advantage, the multi-domain collaborative fusion specifically includes: constructing an inter-domain complementary enhancement matrix ,in Represents the enhanced weight of domain i features on domain j features, through calculate, and is the quality evaluation value of sub-band i, is the feature correlation between sub-bands; calculate the adaptive fusion function ,in The processing results for each domain are: is a dynamic weight; construct edge-aware fusion mask Perform feature consistency correction Calculate the final fusion result .

[0023] Preferably, the variational sharpening operation specifically includes: generating a preliminary estimated image , obtained by low-pass filtering or local statistical characteristics estimation; constructing an improved total variation energy function , where u is the target image, To estimate the ideal image, f is the original image, is the regularization parameter for estimating the ideal image, and is the regularization parameter for the original image; the minimum value of the energy function is solved by split Bregman iteration; the residual r=fu is calculated; based on the residual, adaptive enhancement is performed , where γ is a parameter that is adaptively adjusted according to texture complexity, 0.5≤y≤1.5.

[0024] Preferably, the multi-domain collaborative fusion specifically includes:

[0025] Constructing an inter-domain complementary enhancement matrix ,in express Domain feature pairs Enhanced weights of domain features;

[0026] Compute adaptive fusion function ,in The processing results for each domain are: is the dynamic weight;

[0027] Constructing edge-aware fusion masks ;

[0028] Perform feature consistency correction ;

[0029] Calculate the final fusion result .

[0030] Preferably, the anisotropic diffusion specifically includes:

[0031] Calculating the structure tensor ;

[0032] Performing eigendecomposition on the structure tensor to obtain eigenvalues λ1, λ2 and eigenvectors v1, v2;

[0033] Based on the eigenvalues and eigenvectors, a diffusion tensor is constructed ;

[0034] in, , , k is the edge preservation coefficient;

[0035] Implementing the Diffusion Equation , obtain the enhanced edge subband.

[0036] Preferably, the variational sharpening operation specifically includes:

[0037] Constructing the total variation energy function ,in is the target image, is the original image, is the regularization parameter;

[0038] Solving the minimum value of the energy function by split Bregman iteration;

[0039] Calculate residuals ;

[0040] Based on the residual, adaptive enhancement is performed , where γ is a parameter that is adaptively adjusted according to texture complexity, 0.5 ≤ γ ≤ 1.5.

[0041] Preferably, the specific calculation method of the no-reference image quality assessment index includes:

[0042] Extracting multi-scale structural feature sets from images ;

[0043] Extracting a set of perceptual features from an image ;

[0044] Constructing a multivariate Gaussian model based on natural image statistical models ;

[0045] Calculate the Mahalanobis distance between the test image features and the Gaussian model ;

[0046] in, is the merged eigenvector, is the mean vector of natural image features, is the covariance matrix of natural image features.

[0047] An image effect enhancement system, comprising:

[0048] The content analysis and quantization module is used to determine the optimal quantization level based on the local entropy and global contrast of the image, determine the quantization parameter based on the regional complexity index, and obtain the quantized image and regional complexity map;

[0049] The multi-scale domain conversion module is used to decompose the quantized image into smooth sub-bands, edge sub-bands, and texture sub-bands, extract the main feature vectors of each sub-band, and construct a sub-band feature map;

[0050] A feature enhancement processing module for applying adaptive histogram equalization to smooth subbands, anisotropic diffusion to edge subbands, nonlinear sharpening to texture subbands, and performing multi-scale prediction residual optimization;

[0051] The quality assessment and optimization module is used to calculate the quality assessment index of the reference-free image, construct the regional importance weight map, and dynamically adjust the enhancement parameters based on the assessment results;

[0052] The collaborative fusion and output module is used to construct the inter-domain complementary enhancement matrix, apply the edge-aware fusion mask, perform feature consistency correction, and generate the final enhanced image.

[0053] The present invention has the following beneficial effects:

[0054] 1. The adaptive quantization mechanism improves the adaptability to different types of images and retains more original image details;

[0055] 2. Multi-scale domain decomposition makes the enhancement process more targeted, enabling the most suitable enhancement strategy to be adopted for different types of regions;

[0056] 3. Perceptual quality assessment and feedback optimization enable dynamic adjustment of enhancement parameters to ensure optimal enhancement effects;

[0057] 4. Multi-domain collaborative integration effectively combines the advantages of each domain processing and avoids the limitations of single domain processing;

[0058] 5. The processing efficiency is significantly improved, the processing speed is increased by 3-5 times under the same hardware conditions, and the energy consumption is reduced by about 60%;

[0059] 6. The enhancement effect is more natural, effectively suppressing noise while retaining details, and the visual quality is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is an overall flow chart of an image effect enhancement method of the present invention;

[0061] Figure 2 A flowchart of content-aware adaptive quantization according to the present invention;

[0062] Figure 3 Schematic diagram of multi-scale domain analysis and decomposition of the present invention;

[0063] Figure 4 Flowchart of adaptive feature enhancement and reconstruction of the present invention;

[0064] Figure 5 A schematic diagram of the perceptual quality assessment and feedback optimization of the present invention;

[0065] Figure 6 This is a flow chart of multi-domain collaborative fusion of the present invention;

[0066] Figure 7 This is a structural diagram of an image effect enhancement system according to the present invention;

[0067] Figure 8 The following is a comparison chart of the enhancement effects of the present invention in different application scenarios. DETAILED DESCRIPTION

[0068] Please refer to the attached Figure 1-8 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] like Figure 1 As shown, the present invention provides an image effect enhancement method, comprising the following steps:

[0070] Obtain an image to be processed I;

[0071] Perform content-aware adaptive quantization to obtain the quantized image I_quant and the regional complexity map RC_map. This step determines the optimal quantization level based on the local entropy and global contrast of the image, and determines the quantization parameters for different regions based on the regional complexity index;

[0072] Perform multi-scale domain analysis and decomposition to obtain a multi-scale, multi-directional sub-band set {BS, BE, BT}. This step decomposes the quantized image into smooth sub-bands, edge sub-bands, and texture sub-bands, extracts the main eigenvectors of each sub-band, and constructs a sub-band feature map.

[0073] Perform adaptive feature enhancement and reconstruction to obtain the enhanced subband set {B'S, B'E, B'T}. This step applies adaptive histogram equalization to the smooth subbands, anisotropic diffusion to the edge subbands, nonlinear sharpening to the texture subbands, and performs multi-scale prediction residual optimization.

[0074] Perform perceptual quality assessment and feedback optimization to obtain the optimization parameters P' and the regional importance weight map W. This step calculates the no-reference image quality assessment index and dynamically adjusts the enhancement parameters based on the assessment results.

[0075] Perform multi-domain collaborative fusion to obtain the final enhanced image I''. This step constructs the inter-domain complementary enhancement matrix, applies the edge-aware fusion mask, and performs feature consistency correction.

[0076] The overall process of the method in this invention forms a complete closed-loop system, with each step closely connected, data flow clear, and working together to achieve high-quality image enhancement. Compared with traditional methods, this method introduces innovative mechanisms such as content awareness, multi-domain collaboration, and quality feedback, significantly improving enhancement effects and processing efficiency.

[0077] like Figure 2 As shown, this embodiment describes in detail a method for implementing content-aware adaptive quantization, specifically including:

[0078] Obtain the image's local entropy H_local and global contrast C_global. For each pixel (x, y) in the input image I, calculate the local entropy H_local(x, y) and the global contrast C_global. Determine the quantization level n based on the local entropy and global contrast. When H_local is high and C_global is low, choose a larger r value (e.g., 7-8) to preserve more detail. When H_local is low and C_global is high, choose a smaller n value (e.g., 4-5) to emphasize key structures.

[0079] Ensure 4 Calculate the regional complexity index RC. Preferably, the regional complexity index is calculated by the following formula:

[0080] ,

[0081] Among them, EdgeDensity is the edge density, which indicates the proportion of edge pixels in the image area; GradVariance is the gradient variance, which indicates the degree of change of pixel gradients in the area; MeanIntensity is the average brightness, which indicates the average value of pixel brightness in the area;

[0082] is the resolution correction factor, expressed as , W and H are the width and height of the current image, and The reference standard resolution (usually set to 512×512). This resolution normalization mechanism ensures the consistency of the RC index across different resolutions, preventing high-resolution images from incorrectly obtaining an overly high complexity assessment. The regional complexity index RC reflects the complexity of the image region, with larger values indicating more complex regions.

[0083] Based on the regional complexity index RC, an adaptive weight w(x,y) is generated. The RC value is converted into an adaptive weight through a mapping function, and complex regions are given higher weights, ensuring more refined quantization of these regions.

[0084] Perform quantization operation to obtain quantized image and regional complexity map. Quantization operation is implemented by the following formula:

[0085]

[0086] (x, y) is the pixel value of the original image at point (x, y), and are the minimum and maximum pixel values of the image, n is the quantization level, and W(X,V) is the adaptive weight.

[0087] This embodiment uses an adaptive quantization mechanism to dynamically adjust quantization parameters based on image content characteristics, achieving differentiated processing for different image regions and effectively preserving image detail, laying a solid foundation for subsequent processing. Compared to traditional fixed ternary quantization, adaptive quantization improves information retention by 45-70%, significantly improving the enhancement effect.

[0088] like Figure 3 As shown, this embodiment describes in detail the implementation method of multi-scale domain analysis and decomposition, which specifically includes:

[0089] The quantized image is decomposed into multi-scale, multi-directional sub-bands using a wavelet-directional filter combination transform (WDFB). WDFB combines the multi-resolution characteristics of the wavelet transform with the directional selectivity of the directional filter, effectively capturing edge and texture features in the image.

[0090] Classify multi-directional subbands into smooth subbands , edge subband and texture subbands The classification is based on the frequency and directional characteristics of the sub-bands. The smooth sub-bands mainly contain low-frequency information, the edge sub-bands mainly contain mid-frequency structural information, and the texture sub-bands mainly contain high-frequency details.

[0091] Smooth subband , edge subband and texture subbands Perform principal component analysis separately to extract the main eigenvectors For sub-band , extract its main eigenvectors through principal component analysis:

[0092] ,

[0093] Among them, PCA represents the principal component analysis operation, Indicates subband The main eigenvector of

[0094] Calculate the feature correlation between sub-bands The feature correlation is calculated by the following formula:

[0095] ,

[0096] in, represents the correlation calculation function, and Sub-band and The feature correlation reflects the degree of information correlation between different sub-bands and provides a basis for subsequent collaborative processing;

[0097] Based on the eigenvectors and feature correlations, a sub-band feature map SFM is constructed. The sub-band feature map contains the eigenvectors, statistical characteristics, and correlation information of each sub-band:

[0098] ,

[0099] in, Sub-band The skewness coefficient reflects the asymmetry of the sub-band pixel distribution.

[0100] This embodiment uses multi-scale domain analysis and decomposition to decompose the image into sub-bands with different semantic characteristics, enabling the enhancement process to adopt the most appropriate enhancement strategy for different types of regions. Compared with traditional single-domain processing, multi-domain decomposition provides richer feature expression, laying the foundation for high-quality enhancement.

[0101] like Figure 4 As shown, this embodiment describes in detail the implementation method of adaptive feature enhancement and reconstruction, which specifically includes:

[0102] Smooth subband Applying improved adaptive histogram equalization CLAHE ( , ). Unlike traditional histogram equalization, CLAHE performs equalization in local areas, effectively avoiding artifacts caused by over-enhancement. Parameters Adaptive adjustment based on local contrast:

[0103] ,

[0104] in, is the base parameter value (usually 2.0), is the adjustment factor (usually 0.5), is the contrast value of the local area. Here, Controls the strength of histogram equalization, so that areas with low contrast get stronger enhancement;

[0105] Edge subband Apply anisotropic diffusion AnisoDiff(BE,k,n). Anisotropic diffusion preserves edges while smoothing noise. The parameter k is the edge preservation coefficient, which controls the degree of edge preservation during the diffusion process; η is the diffusion guidance factor, which controls the diffusion direction.

[0106] Texture subband Apply nonlinear sharpening TFS ( ). Nonlinear sharpening is achieved by the following formula:

[0107] ,

[0108] in, is the adaptive gain, (·) is the variational sharpening operation, which enhances texture details by solving the variational problem;

[0109] Perform multi-scale prediction residual optimization The multi-scale prediction residual optimization is achieved through the following formula:

[0110] ,

[0111] in, is the enhanced subband, is the adjustment coefficient, is a prediction function that predicts the residual information of the current scale based on the adjacent scale subbands. Multi-scale prediction residual optimization utilizes the correlation between different scales to improve the consistency of detail enhancement;

[0112] Based on the enhanced subband, an inverse transform reconstruction operation is performed. The inverse transform is implemented by the following formula:

[0113] ,

[0114] in, represents the inverse transform operation of the wavelet-directional filter combination, which reconstructs the enhanced subband into the enhanced image I'.

[0115] This embodiment achieves precise enhancement of different image components by designing differentiated feature enhancement strategies for different sub-band types. A multi-scale prediction residual optimization mechanism further improves cross-scale feature consistency, resulting in a more natural and harmonious enhancement effect. Compared to traditional single enhancement strategies, differentiated enhancement improves edge preservation accuracy by 35% and noise suppression by 28%.

[0116] like Figure 5 As shown, this embodiment details the implementation method of perceptual quality assessment and feedback optimization, specifically including:

[0117] Calculate the No-Reference Image Quality Evaluation Index (NIQE). NIQE is based on a natural image statistical model and comprehensively considers factors such as contrast, clarity, naturalness, and artifact suppression to evaluate the visual quality of the enhanced image.

[0118] Construct the regional importance weight map W(x,y). The regional importance weight map is calculated using the following formula:

[0119] ,

[0120] Where Saliency(x,y) is the saliency value at point (x,y), and InfoGain(x,y) is the information gain value at point (x,y). The regional importance weight map guides the enhancement process to prioritize the quality of visually important regions. Based on the evaluation index and weight map, an evaluation-parameter mapping relationship is constructed. The evaluation-parameter mapping relationship is implemented using the following formula:

[0121] ,

[0122] Where P is the current enhancement parameter, P' is the optimized parameter, Fopt is the optimization function, and ∇NIQE is the gradient of NIQE with respect to parameter P. The quality assessment results are converted into parameter adjustment strategies through the assessment-parameter mapping relationship. The parameters are then iteratively optimized using stochastic gradient descent. Parameter optimization is achieved using the following formula:

[0123] ,

[0124] Where P(t) and P(t+1) are the parameters for the t-th and t+1-th iterations respectively, η is the learning rate, and ∇Cost is the gradient of the cost function. The cost function Cost is defined as:

[0125] ,

[0126] Among them, NIQE(P) is the no-reference quality evaluation score (normalized to between 0 and 1), and the higher the value, the better the quality; Artifact(P) is the artifact evaluation function, which is quantified by the detection of local statistical characteristics anomalies, and the lower the value, the fewer artifacts; Detail(P) is the detail preservation evaluation function, which is quantified by the high-frequency information retention rate, and the lower the value, the better the detail preservation; α, β, and γ are weight coefficients for balancing various indicators, and the value is generally set to a=0.5, The iterative optimization process continues until the termination condition is met: |NIQE(t+1) - NIQE(t)| < or the maximum number of iterations is reached.

[0127] This embodiment implements a perceptual quality assessment and feedback optimization mechanism to dynamically adjust and optimize enhancement parameters, ensuring optimal enhancement results. Compared to traditional fixed-parameter methods, this dynamic optimization mechanism improves enhancement quality by 25-40%, enabling the system to adapt to the processing needs of different image types.

[0128] like Figure 6 As shown, this embodiment details the implementation method of multi-domain collaborative integration, which specifically includes:

[0129] Construct the inter-domain complementary enhancement matrix CEM. The inter-domain complementary enhancement matrix contains the enhancement weight relationship between each domain:

[0130] ,

[0131] in, These weights are automatically calculated by sub-band quality assessment and feature correlation:

[0132] Calculate the adaptive fusion function. The adaptive fusion function is implemented by the following formula:

[0133] ,

[0134] in, and is the quality assessment value of sub-band i and j, is the inter-subband feature correlation. The weight calculation process includes: first, initializing the weight matrix to the identity matrix (self-weight is 1, complementary weight is 0); then, adjusting it based on the subband quality assessment value, so that higher-quality subbands have a greater enhancement contribution to lower-quality subbands; finally, performing weight normalization to ensure that the sum of each row weight is 1. By using the inter-domain complementary enhancement matrix, the system can fully utilize the complementarity of the processing results of different domains; and calculating the adaptive fusion function. The adaptive fusion function is implemented using the following formula:

[0135] ,

[0136] Where Di(x,y) is the value of the processing result of the i-th domain at the point (x,y), and ωi is the dynamic weight, which is dynamically adjusted according to the quality and reliability of the processing results of each domain;

[0137] Construct edge-aware fusion mask. The edge-aware fusion mask is implemented by the following formula:

[0138] ,

[0139] Where EdgeMap(x,y) is the edge strength at point (x,y), and k is the adjustment coefficient, which is usually 5-10. The edge-aware fusion mask ensures seamless transitions in edge areas and avoids edge artifacts during the fusion process.

[0140] Perform feature consistency correction. Feature consistency correction is achieved through the following formula:

[0141] ,

[0142] Where λ is the correction coefficient, which is usually set to 0.2-0.5. Feature consistency correction ensures that the processing results of different domains maintain feature consistency during the fusion process to avoid fusion artifacts;

[0143] Calculate the final fusion result. The final fusion result is achieved through the following formula:

[0144] .

[0145] Where I' is the final enhanced image.

[0146] This embodiment utilizes a multi-domain collaborative fusion strategy to effectively combine the strengths of each domain while avoiding the limitations of single-domain processing. Feature consistency correction and edge-aware fusion masks ensure a natural and harmonious fusion process, effectively suppressing fusion artifacts. Experimental results show that multi-domain collaborative fusion improves subjective scores by 38% and artifact suppression by over 40%.

[0147] like Figure 7 As shown, the present invention also provides an image effect enhancement system, comprising:

[0148] Content analysis and quantization module 1, used to determine the optimal quantization level based on the local entropy and global contrast of the image, determine the quantization parameter based on the regional complexity index, and obtain the quantized image and regional complexity map;

[0149] Multi-scale domain conversion module 2 is used to decompose the quantized image into smooth sub-bands, edge sub-bands and texture sub-bands, extract the main feature vectors of each sub-band, and construct a sub-band feature map;

[0150] Feature enhancement processing module 3, used to apply adaptive histogram equalization to the smooth sub-band, anisotropic diffusion to the edge sub-band, nonlinear sharpening to the texture sub-band, and perform multi-scale prediction residual optimization;

[0151] Quality assessment and optimization module 4, used to calculate the no-reference image quality assessment index, construct the region importance weight map, and dynamically adjust the enhancement parameters based on the assessment results;

[0152] The collaborative fusion and output module 5 is used to construct the inter-domain complementary enhancement matrix, apply the edge-aware fusion mask, perform feature consistency correction, and generate the final enhanced image.

[0153] The system provided by this embodiment has a clear structure and reasonable module division. The modules work together to achieve high-quality image enhancement. The system implements innovative mechanisms such as adaptive processing, multi-domain collaboration, and quality feedback, and has strong adaptability and robustness.

[0154] Each functional module can be implemented using software, hardware, or a combination of both, and can be deployed on personal computers, servers, embedded devices, or dedicated hardware accelerators. The system supports multi-threaded parallel processing and GPU acceleration, meeting the processing speed requirements of different application scenarios.

[0155] This embodiment describes in detail the implementation method of anisotropic diffusion, which is a specific implementation of the edge sub-band enhancement strategy in Example 4, and specifically includes:

[0156] Calculate the structure tensor J. The structure tensor is calculated by the following formula:

[0157] ,

[0158] in, is the image gradient, The structure tensor J describes the directional structure information of the local area of the image;

[0159] Perform eigendecomposition on the structure tensor to obtain eigenvalues 、 and eigenvectors 、 The eigenvalue represents the structural strength along the direction of the eigenvector, and the eigenvector represents the main direction of the local area;

[0160] Construct diffusion tensor based on eigenvalues and eigenvectors The diffusion tensor is constructed using the following formula:

[0161] ,

[0162] in, and is the diffusion coefficient, which is calculated by the following formula:

[0163] ,

[0164] ,

[0165] in, is the edge preservation coefficient, which controls the degree of edge preservation during the diffusion process. When the value is smaller, the edge preservation effect is stronger; when Larger values result in a stronger smoothing effect. The value range is 10-30;

[0166] Execute the diffusion equation to obtain the enhanced edge subband. The diffusion equation is implemented by the following formula:

[0167] ,

[0168] in, represents the divergence operation, The diffusion equation is solved by numerical iteration method, and the number of iterations is usually 5-20.

[0169] This embodiment achieves fine enhancement of edge subbands through anisotropic diffusion technology. Compared with traditional isotropic diffusion, anisotropic diffusion can smooth noise while preserving edge structure, improving the naturalness and clarity of the enhancement effect.

[0170] Example 9: Implementation of variational sharp operation

[0171] This embodiment describes in detail the implementation method of the variational sharpening operation and the specific implementation of the texture subband enhancement strategy, which specifically includes:

[0172] Generate preliminary estimate image The initial estimated image can be obtained by:

[0173] ,

[0174] Where LPF represents the low-pass filtering operation, HPF_enhanced represents the enhanced high-pass filtering operation, and δ is the enhancement coefficient, which is usually set to 1.2-1.8. The estimated image generated in this way retains the basic structure of the original image while moderately enhancing the high-frequency details, providing a more ideal target for variational optimization.

[0175] Variational optimization provides a more ideal objective;

[0176] Construct an improved total variation energy function. The improved total variation energy function is defined by the following formula:

[0177]

[0178] Among them, u is the target image, To estimate the ideal image, f is the original image, μ is the regularization parameter for the estimated ideal image, and its value is usually 0.5-1.0; is the regularization parameter for the original image, and its value is usually 0.1-0.3. This improved energy function takes into account both the estimated ideal image and the original image, which can better guide the texture enhancement process;

[0179] The minimum value of the energy function is solved by split Bregman iteration. Split Bregman iteration is an efficient algorithm for solving variational problems. It decomposes the original problem into multiple sub-problems by introducing auxiliary variables.

[0180] Calculate residuals The residual is calculated using the following formula:

[0181] ,

[0182] residual Contains the original image The high-frequency details that are smoothed out;

[0183] Based on the residual, adaptive enhancement is performed. Adaptive enhancement is achieved by the following formula:

[0184] ,

[0185] in, express The symbol, express The absolute value of is a parameter that is adaptively adjusted according to the texture complexity. When the texture complexity is high, A smaller value (such as 0.5-0.8) will give a stronger enhancement effect; when the texture complexity is low, A larger value (such as 1.2-1.5) will give a milder enhancement effect.

[0186] This embodiment achieves fine-grained enhancement of texture subbands through an improved variational sharpening operation. Compared to traditional sharpening methods, this improved variational sharpening better preserves image structure while enhancing texture details, avoiding noise amplification and ringing artifacts caused by over-sharpening. By introducing an estimated ideal image as a reference, it overcomes the limitations of directly using low-quality original images as a reference, further improving the quality and naturalness of the enhancement effect.

[0187] This embodiment describes in detail a method for calculating a no-reference image quality assessment index, specifically including:

[0188] Extract the multi-scale structural feature set S of the image. The multi-scale structural features are obtained through Gaussian pyramid decomposition and gradient statistics, reflecting the structural information of the image such as edges and textures;

[0189] Extract the perceptual feature set P of the image. Perceptual features include local contrast, brightness distribution, color saturation, etc., which reflect the visual perception characteristics of the image;

[0190] Based on the natural image statistical model, a multivariate Gaussian model is constructed. The multivariate Gaussian model is represented by the following parameters:

[0191] ,

[0192] in, is the mean vector, is the covariance matrix. The multivariate Gaussian model describes the distribution characteristics of high-quality natural image features;

[0193] Calculate the Mahalanobis distance between the test image features and the Gaussian model. The Mahalanobis distance is calculated using the following formula:

[0194] ,

[0195] in, is the feature vector of the test image, which is a combination of structural features and perceptual features: =[S,P]. Mahalanobis distance reflects the degree of difference between the test image and the statistical model of natural images. The smaller the distance, the higher the quality.

[0196] This embodiment uses a no-reference image quality assessment index to objectively evaluate enhanced image quality, providing a reliable basis for parameter optimization. Compared to traditional full-reference evaluation metrics, the no-reference evaluation index does not require the original reference image and is more suitable for practical application scenarios. Experimental results show that the consistency of this evaluation index with human subjective evaluation exceeds 85%.

[0197] This embodiment verifies the effectiveness of the method and system of the present invention through actual application cases. Figure 8 As shown, in different application scenarios, the method of the present invention has obvious advantages over the prior art:

[0198] In medical imaging scenarios, the method of the present invention improves the visibility of details in low-dose CT images, increasing the signal-to-noise ratio by 4.2dB while maintaining key diagnostic features, reducing the misdiagnosis rate of radiologists by 12.5% and improving diagnostic efficiency by 23%.

[0199] In video surveillance scenarios, the method of the present invention improves target detection rates by 32% and reduces false alarm rates by 18% under low-light conditions. It also significantly improves real-time processing capabilities and increases the number of cameras supported by the same hardware by 75%.

[0200] In mobile device imaging scenarios, the method of the present invention improves photo quality, increases user satisfaction by 28%, and reduces processing power consumption by 35% for the same processing task.

[0201] In satellite remote sensing scenarios, the method of the present invention improves the ability to identify details, increases the accuracy of object classification by 15-20%, increases the degree of automation in the processing flow, and reduces the need for manual intervention by 65%.

[0202] Through application tests in different scenarios, the versatility and effectiveness of the method of the present invention have been verified. The present invention not only outperforms the existing technology in objective indicators, but also shows obvious advantages in subjective visual quality and practical application effects.

[0203] The multi-domain collaborative adaptive image enhancement method and system of the present invention have broad industrial application value:

[0204] 1. In the field of medical imaging, it can be used to improve the quality of low-dose CT, X-ray, ultrasound and other medical images, helping doctors make more accurate diagnoses;

[0205] 2. In the field of video surveillance, it can be used to improve the quality of surveillance videos in harsh conditions such as low light, rain and fog, and increase target detection and recognition rates;

[0206] 3. In the field of mobile devices, it can be integrated into the image processing flow of devices such as smartphones and tablets to enhance the user's photography experience;

[0207] 4. In the field of satellite remote sensing, it can be used to improve the clarity and usability of satellite images and enhance the accuracy of object recognition and classification;

[0208] 5. In the field of industrial inspection, it can be used to improve the quality of images collected by industrial cameras and increase the accuracy of defect detection.

[0209] In general, the present invention provides an image enhancement solution with high adaptability, high efficiency and high quality, which is suitable for various image processing application scenarios and has significant technical innovation and practical value.

[0210] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for enhancing image effects, characterized in that: include: Get the image to be processed; Perform content-aware adaptive quantization to obtain a quantized image and a regional complexity map, including: determining the optimal quantization level based on the local entropy and global contrast of the image; and determining quantization parameters for different regions based on the regional complexity index; Perform multi-scale domain analysis and decomposition to obtain a multi-scale and multi-directional sub-band set, including: decomposing the quantized image into smooth sub-bands, edge sub-bands, and texture sub-bands; extracting the main eigenvectors of each sub-band and constructing a sub-band feature map; Perform adaptive feature enhancement and reconstruction to obtain a set of enhanced subbands, including: applying adaptive histogram equalization to smooth subbands; applying anisotropic diffusion to edge subbands; applying nonlinear sharpening to texture subbands; and performing multi-scale prediction residual optimization. Perform perceptual quality assessment and feedback optimization to obtain optimization parameters and regional importance weight maps, including: calculating the no-reference image quality assessment index; dynamically adjusting enhancement parameters based on the assessment results; Perform multi-domain collaborative fusion to obtain the final enhanced image, including: constructing inter-domain complementary enhancement matrices; applying edge-aware fusion masks; and performing feature consistency correction.

2. The method according to claim 1, characterized in that The content-aware adaptive quantization specifically includes: obtaining local entropy and global contrast of the image; determining a quantization level n based on the local entropy and global contrast of the image, where 4≤n≦8; and calculating a regional complexity index. Among them, EdgeDensity is the edge density, which indicates the proportion of edge pixels in the image area; GradVariance is the gradient variance, which indicates the degree of change of pixel gradients in the area; MeanIntensity is the average brightness, which indicates the average brightness of pixels in the area; is the resolution correction factor, expressed as , W and H are the width and height of the current image, and is the reference standard resolution; based on the regional complexity index, an adaptive weight w(x,y) is generated; and a quantization operation is performed: ,Get quantized images and regional complexity maps.

3. The method according to claim 1, characterized in that The perceptual quality assessment and feedback optimization specifically include: calculating the no-reference image quality evaluation index NIQE=f(contrast, clarity, naturalness, artifact suppression); constructing a regional importance weight map ; Based on the evaluation index and weight map, construct the evaluation-parameter mapping relationship Iterative optimization of processing parameters via stochastic gradient descent ,in, P is the comprehensive cost function, NIQE(P) is the no-reference quality evaluation score (normalized to between 0 and 1), Artifact(P) is the artifact evaluation function, Detail(P) is the detail preservation evaluation function, and α, β, and γ are the weight coefficients for balancing various indicators. Iterative optimization continues until or the maximum number of iterations is reached.

4. The method according to claim 1, wherein The multi-domain collaborative fusion specifically includes: constructing an inter-domain complementary enhancement matrix , in Represents the enhanced weight of domain i features on domain j features, through calculate, and is the quality evaluation value of sub-band i, is the feature correlation between sub-bands; calculate the adaptive fusion function ,in The processing results for each domain are: is a dynamic weight; construct edge-aware fusion mask Perform feature consistency correction Calculate the final fusion result .

5. The method according to claim 4, characterized in that The variational sharpening operation specifically includes: generating a preliminary estimated image , obtained by low-pass filtering or local statistical characteristics estimation; constructing an improved total variation energy function , where u is the target image, To estimate the ideal image, f is the original image, is the regularization parameter for estimating the ideal image, and is the regularization parameter for the original image; the minimum value of the energy function is solved by split Bregman iteration; the residual r=fu is calculated; based on the residual, adaptive enhancement is performed , where γ is a parameter that is adaptively adjusted according to texture complexity, 0.5≤y≤1.

5.

6. The method according to claim 1, characterized in that The multi-domain collaborative integration specifically includes: Constructing an inter-domain complementary enhancement matrix ,in express Domain feature pairs Enhanced weights of domain features; Compute adaptive fusion function ,in The processing results for each domain are: is the dynamic weight; Constructing edge-aware fusion masks ; Perform feature consistency correction ; Calculate the final fusion result .

7. The method according to claim 4, characterized in that The anisotropic diffusion specifically includes: Calculating the structure tensor ; Performing eigendecomposition on the structure tensor to obtain eigenvalues λ1, λ2 and eigenvectors v1, v2; Based on the eigenvalues and eigenvectors, a diffusion tensor is constructed ; in, , , k is the edge preservation coefficient; Implementing the Diffusion Equation , obtain the enhanced edge subband.

8. The method according to claim 4, characterized in that The variational sharpening operation specifically includes: Constructing the total variation energy function ,in is the target image, is the original image, is the regularization parameter; Solving the minimum value of the energy function by split Bregman iteration; Calculate residuals ; Based on the residual, adaptive enhancement is performed , where γ is a parameter that is adaptively adjusted according to texture complexity, 0.5 ≤ γ ≤ 1.

5.

9. The method according to claim 5, characterized in that The specific calculation method of the no-reference image quality assessment index includes: Extracting multi-scale structural feature sets from images ; Extracting a set of perceptual features from an image ; Constructing a multivariate Gaussian model based on natural image statistical models ; Calculate the Mahalanobis distance between the test image features and the Gaussian model ; in, is the merged eigenvector, is the mean vector of natural image features, is the covariance matrix of natural image features.

10. An image enhancement system for executing the method according to any one of claims 1 to 9, characterized in that: include: The content analysis and quantization module is used to determine the optimal quantization level based on the local entropy and global contrast of the image, determine the quantization parameter based on the regional complexity index, and obtain the quantized image and regional complexity map; The multi-scale domain conversion module is used to decompose the quantized image into smooth sub-bands, edge sub-bands, and texture sub-bands, extract the main feature vectors of each sub-band, and construct a sub-band feature map; A feature enhancement processing module for applying adaptive histogram equalization to smooth subbands, anisotropic diffusion to edge subbands, nonlinear sharpening to texture subbands, and performing multi-scale prediction residual optimization; The quality assessment and optimization module is used to calculate the quality assessment index of the reference-free image, construct the regional importance weight map, and dynamically adjust the enhancement parameters based on the assessment results; The collaborative fusion and output module is used to construct the inter-domain complementary enhancement matrix, apply the edge-aware fusion mask, perform feature consistency correction, and generate the final enhanced image.

Images