Image enhancement method and system based on deep learning

Abstract

The present invention discloses a method and system for image enhancement based on deep learning, comprising the following steps: obtaining an image pixel matrix, partitioning the frequency according to brightness difference, gradient direction, and texture density, extracting multi-region features and generating a weighted response map, performing scale decomposition to extract multi-scale features, fusing residual maps to enhance details and suppress redundancy, calling multi-map parameter brightness callbacks, and outputting an enhanced image. In the present invention, accurate separation of high- and low-frequency regions is achieved through regional frequency attribute division, avoiding blurring and over-enhancement caused by unified processing. Different strategies are adopted for different regions to enhance high-frequency details and suppress low-frequency redundancy. Feature response weight maps are weighted and normalized to enhance local texture and contrast. Scale decomposition combined with multi-scale fusion enhances naturalness and avoids visual discomfort. Edge consistency matching optimizes brightness callbacks to enhance edge clarity and overall visual performance.

Description

Technical Field

[0001] The present invention relates to the field of non-spatial domain filtering technology, and in particular to an image enhancement method and system based on deep learning. Background Art

[0002] The field of non-spatial domain filtering technology includes technical methods for improving image quality during image processing, especially image enhancement methods. This technical field mainly involves improving the contrast, brightness, clarity and other visual effects of images by processing the non-spatial domain features of images. Non-spatial domain filtering technology generally does not rely on the spatial pixel values ​​of the image, but instead optimizes by analyzing the frequency domain information of the image or the characteristics of other transform domains. In this field, the frequency components, color models, time domain characteristics or other mathematical transformations of the image are usually used as the basis for filtering. This technical field is not only widely used in image enhancement, but also has important significance in medical imaging, satellite image processing, industrial automation and other fields.

[0003] Among them, the image enhancement method refers to improving the image through a specific processing method, thereby improving the visual effect of the image. This method targets the problem of degraded visual quality in the image due to noise, poor lighting conditions, insufficient contrast, etc. The image enhancement method is usually implemented by adjusting the grayscale value of the image, optimizing the contrast, enhancing the color, etc. Specifically, the subject of the patent performs frequency domain filtering on the image in the non-spatial domain and uses frequency component analysis to distinguish image features in different areas. The purpose of enhancing image details is achieved by processing high-frequency and low-frequency information differently. The patented method uses frequency domain transformation technology, such as Fourier transform or wavelet transform, to convert the image from the spatial domain to the frequency domain, and then optimizes the image performance by selectively filtering and enhancing specific frequency components.

[0004] Existing technologies lack the ability to fine-tune the characteristics of image regions and dynamically adjust them. They often uniformly process the frequency components of the entire image, resulting in inaccurate processing of image details and redundant information. In low-frequency areas, noise suppression is not ideal, while high-frequency areas are prone to over-enhancement, causing distorted or unnatural details. Traditional methods are relatively simple in processing edge areas, which can easily cause blurring and affect the overall image clarity. In practical applications, especially in the fields of medical and satellite imaging, insufficient or distorted image details can lead to the loss of key details, affecting the accuracy of the analysis results. These deficiencies stem from the fact that existing technologies fail to fully consider regional differences and lack more flexible means of detail enhancement. Summary of the Invention

[0005] The purpose of the present invention is to address the shortcomings of the prior art and to propose an image enhancement method and system based on deep learning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an image enhancement method based on deep learning, comprising the following steps:

[0007] S1: Obtain the input image pixel matrix, determine the regional frequency attribute based on the brightness difference, gradient direction change value, local change trend intensity and texture density distribution, divide it into high-frequency area and low-frequency area according to the edge response threshold, and obtain the image frequency partition label;

[0008] S2: Based on the image frequency partition label, extract the edge intensity value, texture fineness value, contrast value, gradient direction change value, spectrum balance value and color distribution concentration of multiple regions, perform nonlinear cross integration operation, assign the results to the corresponding region, and obtain the regional feature response weight map;

[0009] S3: Decomposing the input image according to the regional feature response weight map, extracting a multi-scale edge-guided feature map, a texture map, and a brightness map, adjusting the residual response values ​​of the multiple maps according to the weights, enhancing the details of the high-frequency region and suppressing the redundancy of the low-frequency region, fusing all the scale maps and constructing a mapping structure to obtain an image enhancement residual map;

[0010] S4: Based on the image enhancement residual map, call the scale map brightness mean, edge direction distribution map and texture contrast map, match the area according to the edge consistency coefficient and perform brightness callback to obtain a multi-scale enhanced fusion image.

[0011] As a further solution of the present invention, the image frequency partition label is specifically a high-frequency area and a low-frequency area, the regional feature response weight map includes multi-region edge intensity values, texture fineness values, and contrast values, the image enhancement residual mapping map is specifically a multi-scale edge-guided feature map, texture mapping map, and brightness mapping map after scale decomposition, and the multi-scale enhanced fusion image includes the scale map brightness mean, edge direction distribution map, and texture contrast map.

[0012] As a further solution of the present invention, the steps of obtaining the image frequency partition label are specifically as follows:

[0013] S101: Based on the input image pixel matrix, calculate the grayscale difference and compare it with the brightness difference threshold, count and normalize the number of pixels with significant brightness difference, and obtain the brightness difference value;

[0014] S102: Call the brightness difference value, and use the formula:

[0015]

[0016] The regional gradient response value is obtained by calculation and superimposed with the distribution of brightness difference density to obtain the frequency mutation factor value;

[0017] Among them, F Δg Represents the regional gradient response value, Δg a Represents the gradient direction change value of the a-th pixel pair, G a Indicates the local change trend strength of the point, H a is the point gradient change density value, B a is the brightness difference of the point neighborhood, and N represents the number of pixel pairs participating in the statistics;

[0018] S103: comparing the frequency mutation factor value with the texture density distribution and the edge response threshold, determining the frequency attribute of the region, and generating an image frequency partition label.

[0019] As a further solution of the present invention, the step of obtaining the regional feature response weight map is specifically as follows:

[0020] S201: extracting edge intensity values, texture fineness values, and contrast values ​​of multiple regions based on the image frequency partition labels, normalizing them according to the maximum and minimum values, and generating multi-region standardized image feature values;

[0021] S202: Calling the multi-region standardized image feature values, collecting gradient direction change values, spectrum balance values ​​and color distribution concentration, performing nonlinear cross integration operations, using the formula:

[0022]

[0023] Calculate and obtain regional characteristic response values, adjust the attribute response ratio within the region, and generate regional characteristic response distribution values;

[0024] Among them, RC b represents the characteristic response value of region b, R b represents the normalized edge strength value of the bth region, F b represents the normalized texture fineness value of the b-th region, O b represents the normalized contrast value of the b-th region, G b Indicates the gradient direction change value of the b-th region, W b Indicates the spectrum balance value of the b-th area, Z b represents the color distribution concentration of the b-th area,

[0025] ∑(W b +Z b ) represents the joint summation of spectrum and color;

[0026] S203: Mapping a multi-region response value grid to an image space position according to the regional feature response distribution value, establishing a regional weight distribution matrix, and obtaining a regional feature response weight map.

[0027] As a further solution of the present invention, the steps of obtaining the image enhancement residual map are specifically as follows:

[0028] S301: performing scale decomposition on the input image according to the regional feature response weight map, extracting the edge-guided feature map, the texture map, and the brightness map, performing regional feature response weight comparison, extracting high-frequency regional features, and generating multi-scale regional response feature values;

[0029] S302: Call the multi-scale regional response characteristic value, calculate the response difference value, construct the characteristic sensitivity coefficient, and adjust the residual response with the brightness map difference value, using the formula:

[0030]

[0031] Calculate and obtain the weight-adjusted residual structure;

[0032] Among them, W res Represents the weight-adjusted residual structure, ΔN c represents the intensity difference of the edge-guided feature map at the cth scale, ΔG c Represents the structural difference of the texture map at the cth scale, Y c Represents the grayscale distribution value of the brightness map at the cth scale, M c Indicates the characteristic sensitivity coefficient, max c Indicates the maximum value operation among all scales;

[0033] S303: Adjust the residual structure amount according to the weight, fuse the residual signals of all scale maps, construct a multi-scale structural relationship of image elements, and generate an image enhancement residual map.

[0034] As a further solution of the present invention, the steps of acquiring the multi-scale enhanced fusion image are specifically as follows:

[0035] S401: Based on the image enhancement residual map, call the scalogram brightness mean and edge direction distribution map, calculate the brightness offset difference and the edge direction angle value, screen the area that meets the edge consistency threshold condition, and obtain the edge consistency area distribution coefficient;

[0036] S402: Extract the texture contrast value of the matching area based on the edge consistent area distribution coefficient, and construct the brightness difference and texture contrast difference ratio using the formula:

[0037]

[0038] Calculate the ratio of overall brightness deviation to texture distinctness, reorder regional brightness according to texture features, and obtain a brightness callback reference sequence;

[0039] Among them, V represents the ratio of overall brightness deviation to texture distinctness, u represents the number of pixels involved in the calculation in the edge consistent area, and Q d Represents the brightness value of the d-th pixel in the edge consistent area, Indicates the mean brightness of the pixel in the scale map, X d Indicates the texture contrast value corresponding to the d-th pixel in the texture contrast map, max(X d ) represents the maximum value of the texture contrast value corresponding to all participating pixels in the current area, K d,r Represents the brightness difference between the d-th pixel and the adjacent r-th pixel;

[0040] S403: Based on the brightness callback reference sequence, call the matching area brightness information, perform brightness mapping and segment adjustment, integrate texture features and edge direction features, and obtain a multi-scale enhanced fusion image.

[0041] As a further embodiment of the present invention, the method further comprises:

[0042] S5: Based on the multi-scale enhanced fusion image, the edge response difference map, texture consistency offset value and brightness change rate of the original image and the enhanced image are called, and an enhancement amplitude judgment operation is performed on all high-frequency areas in the image. The area with a response amplitude lower than the edge enhancement judgment threshold is eliminated, and the original image value is restored to the area, and a regional enhanced image result map is output;

[0043] The region-enhanced image result diagram is specifically an image after the high-frequency region enhancement amplitude is determined.

[0044] As a further solution of the present invention, the steps of obtaining the region enhanced image result map are specifically as follows:

[0045] S501: Based on the multi-scale enhanced fusion image, call the edge response difference map of the original image and the enhanced image to obtain the regional edge response strength, compare the gradient response value of the difference map at the regional edge position with the set edge enhancement discrimination threshold, and determine whether the region belongs to the high-frequency change region, and generate a high-frequency response interval value;

[0046] S502: Call the high frequency response interval value, combine the texture consistency offset value and the brightness change rate, and use the formula for all areas in the high frequency response interval:

[0047]

[0048] Calculate the enhancement amplitude index, filter the area greater than the edge enhancement discrimination threshold, and obtain the edge enhancement effective coefficient distribution;

[0049] in, Indicates the enhancement amplitude index of the coordinate (e, s) position, Ue,s Indicates the texture consistency offset value of the position, L e,s Indicates the rate of change of brightness, ΔP e,s represents the response value of the edge response difference map, represents the enhancement value of the x-th layer fusion image at the position under the differential scale, and v represents the number of scales involved in the fusion image;

[0050] S503: calling the edge enhancement effective coefficient distribution, identifying the area below the edge enhancement discrimination threshold, restoring the image value to the original image value, and maintaining the enhancement process for the remaining areas to obtain the regional enhanced image result map.

[0051] A deep learning-based image enhancement system, wherein the deep learning-based image enhancement system is used to execute the above-mentioned deep learning-based image enhancement method, and the system comprises:

[0052] The image frequency partition module obtains the input image pixel matrix, extracts the brightness difference, gradient direction change value and local change trend intensity, makes judgments based on the product of the brightness difference and texture density distribution and the gradient direction change value, divides the frequency region and marks the image block type, and generates the image frequency partition label;

[0053] The regional feature mapping module calls the image frequency partition label, extracts the regional edge intensity value, texture fineness value and contrast value, normalizes them and weightedly superimposes them and assigns them to the corresponding image block position to generate a regional feature response weight map;

[0054] The scale residual fusion module calls the regional feature response weight map, obtains the image block content and extracts the multi-scale edge-guided map, texture map and brightness map, performs residual calculation on the three maps according to the image block weight value and performs weighted fusion, and superimposes the fusion value according to the image block position to reconstruct the entire image and generate an image enhancement residual map;

[0055] The edge brightness callback module calls the image enhancement residual map, collects the brightness mean, edge direction distribution map and texture contrast map, determines the regional matching relationship based on the edge direction consistency coefficient and performs brightness linear callback to generate a multi-scale enhanced fusion image;

[0056] The high-frequency detail restoration module calls the multi-scale enhanced fusion image, extracts the edge response difference map between it and the original image, and makes a combined judgment based on the texture consistency offset value and the brightness change rate. It determines that the unsatisfactory area is restored to the original image value and generates a regional enhanced image result map.

[0057] Compared with the prior art, the advantages and positive effects of the present invention are:

[0058] In the present invention, accurate separation of high-frequency and low-frequency regions is achieved through the division of regional frequency attributes, avoiding the blurring and over-enhancement problems caused by overall unified processing. Different processing strategies are adopted for different regions to effectively enhance the details of high-frequency regions and suppress redundant information in low-frequency regions. The feature response weight map enhances the local texture and contrast in the image through weighted superposition and normalization processing, making the details and structure of the image more prominent. In addition, the scale decomposition method is combined with the fusion of multi-scale maps to make the image enhancement effect more natural and avoid the visual discomfort caused by over-processing. The brightness callback is optimized through edge consistency matching, making the details of the edge area clearer, the overall visual effect of the image richer, and enhancing the expressiveness and readability of the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0060] Figure 2 Flowchart of the steps for obtaining the image frequency partition label of the present invention;

[0061] Figure 3 Flowchart of the steps for obtaining the regional characteristic response weight map of the present invention;

[0062] Figure 4 Flowchart of the steps for obtaining the image enhancement residual map of the present invention;

[0063] Figure 5 Flowchart of the steps for acquiring multi-scale enhanced fusion images of the present invention;

[0064] Figure 6 This is a flow chart of the steps for obtaining the regional enhanced image result map of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0066] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0067] Example 1

[0068] See also Figure 1 , the present invention provides a technical solution: an image enhancement method based on deep learning, comprising the following steps:

[0069] S1: Obtain the input image pixel matrix, determine the regional frequency attribute based on the brightness difference, gradient direction change value, local texture density, local change trend intensity and texture density distribution, divide it into high-frequency area and low-frequency area according to the edge response threshold, and obtain the image frequency partition label;

[0070] S2: Based on the image frequency partition label, extract the edge intensity value, texture fineness value, contrast value, gradient direction change value, spectrum balance value and color distribution concentration of multiple regions, perform nonlinear cross integration operation, assign the results to the corresponding region, and obtain the regional feature response weight map;

[0071] S3: Based on the regional feature response weight map, the input image is scaled and decomposed to extract the multi-scale edge-guided feature map, texture map and brightness map. The residual response values ​​of the multiple maps are adjusted according to the weights to enhance the details of the high-frequency area and suppress the redundancy of the low-frequency area. All scale maps are fused and a mapping structure is constructed to obtain the image enhancement residual map.

[0072] S4: Based on the image enhancement residual map, call the scale map brightness mean, edge direction distribution map and texture contrast map, match the area according to the edge consistency coefficient and perform brightness callback to obtain a multi-scale enhanced fusion image;

[0073] S5: Based on the multi-scale enhanced fusion image, the edge response difference map, texture consistency offset value and brightness change rate of the original image and the enhanced image are called, and the enhancement amplitude judgment operation is performed on all high-frequency areas in the image. The areas with response amplitudes lower than the edge enhancement judgment threshold are eliminated, and the original image values ​​are restored to them. The regional enhanced image result map is output.

[0074] The image frequency partition labels are specifically high-frequency areas and low-frequency areas. The regional feature response weight map includes multi-region edge strength values, texture fineness values, and contrast values. The image enhancement residual mapping map is specifically the multi-scale edge-guided feature map, texture mapping map, and brightness mapping map after scale decomposition. The multi-scale enhanced fusion image includes the scale map brightness mean, edge direction distribution map, and texture contrast map. The regional enhanced image result map is specifically the image after the high-frequency region enhancement amplitude judgment.

[0075] See also Figure 2 , the steps to obtain the image frequency partition label are as follows:

[0076] S101: Based on the input image pixel matrix, calculate the grayscale difference and compare it with the brightness difference threshold, count and normalize the number of pixels with significant brightness difference, and obtain the brightness difference value;

[0077] First, the image pixels are read row by row and column by column. The grayscale value of each pixel is calculated using the standard RGB weighted average formula (Gray = 0.299R + 0.587G + 0.114B) to obtain the grayscale value of each pixel, and then a grayscale image matrix is ​​established. For example, for a 4×4 image pixel matrix (Table 1 lists the RGB value of each pixel and the corresponding grayscale calculation result):

[0078] Table 1 Pixel RGB value and corresponding gray value calculation table

[0079] Pixel position R G B Grayscale value (1,1) 60 70 80 69 (1,2) 62 75 83 73 (1,3) 200 205 210 204 (1,4) 198 202 208 201 … … … … …

[0080] As shown in Table 1, the grayscale difference calculation is performed for adjacent pixels in the image grayscale matrix. For example, the grayscale difference between the pixels at positions (1,1) and (1,2) is calculated as: |69-73|=4. Similarly, the grayscale difference between the pixels at positions (1,3) and (1,4) is calculated as:

[0081] |204 - 201| = 3. The above process is repeated for all pairs of adjacent pixels, obtaining all grayscale difference values. Next, the grayscale difference results for each pair of pixels are compared against a brightness difference threshold. The brightness difference threshold is set within the range [0, 255] based on empirical data. Here, the threshold is set to 10 because, in real-world scenarios, a brightness difference threshold of 10 can more accurately distinguish between areas with significant brightness changes and those with less significant changes. For example, the grayscale difference between pixels (1, 1) and (1, 2) is 4, which is less than the threshold of 10 and is therefore considered a non-significant difference. The grayscale difference between pixels (1, 3) and (1, 4) is 3, also considered a non-significant difference. If a point has a grayscale difference of 15, it is considered a significant difference. This process continues for all pairs of adjacent pixels. Then, the number of pixels with significant brightness differences in the entire image is counted and normalized by dividing it by the total number of pixel pairs in the image. For example, if an image contains 500 pixel pairs and the number of pixels with significant brightness differences is 120, then the brightness difference is 120 / 500 = 0.24, which is the "brightness difference" obtained in this step.

[0082] S102: Call the brightness difference value, based on the multi-pixel gradient direction change value, the point gradient change density value and the local change trend intensity, using the formula:

[0083]

[0084] The regional gradient response value is obtained by calculation and superimposed with the distribution of brightness difference density to obtain the frequency mutation factor value;

[0085] Among them, F Δg Represents the regional gradient response value, Δg a Represents the gradient direction change value of the a-th pixel pair, G a Indicates the local change trend strength of the point, H a is the point gradient change density value, B a is the brightness difference of the point neighborhood, and N represents the number of pixel pairs participating in the statistics;

[0086] Based on the "brightness difference" obtained in the previous step, the gradient direction change value Δg of multiple pixel points of the image is collected a , local gradient direction change trend intensity G a , gradient change density value H a And the neighborhood brightness difference B a The specific process of obtaining the above parameters is as follows: for example, 3 groups of pixel pairs are randomly selected in the image pixel area as examples, and the data are as follows: (pixel pair 1: Δg1=5°, G1=0.8, H1=0.2, B1=0.5), (pixel pair 2: Δg2=8°, G2=0.6, H2=0.25, B2=0.45), (pixel pair 3: Δg3=10°, G3=0.7, H3=0.3, B3=0.55). Substitute the above data into the formula to calculate the regional gradient response value F Δg :

[0087]

[0088] The specific process of step-by-step calculation is as follows:

[0089] Pixel-to-pixel calculation:

[0090]

[0091] Calculation of pixel pair 2:

[0092]

[0093] Calculation of pixel pair 3:

[0094]

[0095] Substituting and summing:

[0096]

[0097] Subsequently, the regional gradient response value is superimposed with the brightness difference value, and the final frequency mutation factor value (7.643×0.24=1.834 in this example) is obtained through numerical calculation (such as multiplication and superposition). This result is the "frequency mutation factor value" obtained in this step.

[0098] S103: comparing the edge response threshold with the frequency mutation factor value and the texture density distribution, determining the frequency attribute of the region, and generating an image frequency partition label.

[0099] For example, sampling is performed on a local area, different areas are selected, and their texture density values ​​are calculated. The data obtained is assumed to be: [1.0, 1.5, 2.0, 1.8, 1.2]. The average value is: (1.0+1.5+2.0+1.8+1.2) / 5=1.5, which is used as the texture density benchmark value. Then, by comparing the frequency mutation factor value 1.834 with the texture density baseline value 1.5, the difference between the two is obtained, 1.834-1.5=0.334, and then compared with the edge response threshold, where the edge response threshold is set to 0.3 (in practical applications, it is generally set between [0.1,0.5]). In this example, the difference 0.334 is greater than the set threshold 0.3, so the area is judged to be a high-frequency attribute area. Based on this judgment result, this area is marked as a high-frequency area (labeled as "high frequency"). If it is less than the threshold, it is marked as "low frequency", and the intermediate case is marked as "medium frequency". Finally, the frequency attribute division of the entire image area is achieved one by one, and the frequency partition label is obtained.

[0100] See also Figure 3 , the steps for obtaining the regional feature response weight map are as follows:

[0101] S201: extracting edge intensity values, texture fineness values, and contrast values ​​of multiple regions based on image frequency partition labels, normalizing them according to maximum and minimum values, and generating multi-region standardized image feature values;

[0102] For example, taking a remote sensing image as an actual object, assume that the image is divided into several characteristic regions, where Regions 1, 2, and 3 are selected as case study regions. Within each region, initial extraction of edge strength, texture fineness, and contrast values ​​is performed. Specifically, for edge strength, the Sobel operator is first used to extract edge information in the horizontal and vertical directions. Assuming that the horizontal and vertical gradient values ​​obtained in Region 1 are 6 and 8, respectively, the edge strength value is calculated as the square root of the sum of the squares: Subsequently, the texture fineness value within the region is extracted using the grayscale co-occurrence matrix. Assuming that the texture fineness of region 1 is initially calculated to be 0.65, the contrast is obtained by taking the difference between the maximum and minimum grayscale values ​​of the pixels within the region. The maximum grayscale of the pixels in region 1 is 180, and the minimum grayscale is 60, so the initial contrast value is 180-60=120. The data for the aforementioned regions are summarized in Table 2:

[0103] Table 2. Regional initial image feature value table

[0104] Area Number Edge strength value Texture fineness value Contrast value Area 1 10 0.65 120 Area 2 8 0.55 95 Area 3 6 0.75 130

[0105] As shown in Table 2, the initial eigenvalues ​​of each region are normalized to their maximum and minimum values. Taking the edge intensity value of region 1 as an example, the normalization calculation method is: (actual value - minimum value) / (maximum value - minimum value), that is: (10-6) / (10-6) = 1, the edge intensity value of region 2 is normalized to (8-6) / (10-6) = 0.5, and the edge intensity value of region 3 is 0; the texture fineness value and contrast value are normalized in a similar way, and finally the multi-feature normalized value matrix of each region is obtained, thereby generating the multi-region normalized image eigenvalue.

[0106] S202: Calling multi-region standardized image feature values, collecting gradient direction change values, spectrum balance values ​​and color distribution concentration, and performing nonlinear cross integration operations using the formula:

[0107]

[0108] Calculate and obtain regional characteristic response values, adjust the attribute response ratio within the region, and generate regional characteristic response distribution values;

[0109] Among them, RC b represents the characteristic response value of region b, R b represents the normalized edge strength value of the bth region, F b represents the normalized texture fineness value of the b-th region, O b represents the normalized contrast value of the b-th region, G b Indicates the gradient direction change value of the b-th region, W b Indicates the spectrum balance value of the b-th area, Z b represents the color distribution concentration of the b-th region, ∑(W b +Z b ) represents the joint summation of spectrum and color;

[0110] Call the multi-region standardized image feature values ​​generated above and select the normalized edge intensity value R b , normalized texture fineness value F b , normalized contrast value O b As the initial feature parameter. In addition, it is necessary to further measure and calculate the gradient direction change value G of each area b , spectrum balance value W b , color distribution concentration Z b For example, for region 1, through the gradient direction statistical analysis, if the standard deviation of the pixel gradient direction in the region is measured to be 45°, then the gradient direction change value G b =45; the spectrum balance value is quantitatively evaluated by the uniformity of the regional spectrum curve distribution. Assuming that the spectrum curve fitting goodness in region 1 is 0.85, the spectrum balance value W b=0.85; the color distribution concentration is calculated by the regional RGB color variance. Assuming that the color distribution variance of region 1 is 0.2, the color distribution concentration Z b =0.2. The above parameters for area 1 are listed as follows:

[0111] Normalized edge strength value R b =1, normalized texture fineness value F b =0.5, normalized contrast value O b =0.7, gradient direction change value G b =45, spectrum balance value W b =0.85, color distribution concentration Z b =0.2;

[0112] Substitute the above data into the formula:

[0113]

[0114] The specific operation process is as follows:

[0115] First calculate the numerator:

[0116]

[0117] Then calculate the denominator, which is the sum of the spectrum balance and color concentration of each area. Assuming that the data of all three areas are as shown in Table 2, we have:

[0118] (W1+Z1)+(W2+Z2)+(W3+Z3)=(0.85+0.2)+(0.75+0.3)+(0.65+0.4)=3.15;

[0119] Therefore, the characteristic response value of region 1 is:

[0120]

[0121] Similarly, the characteristic response values ​​of regions 2 and 3 are calculated, and then the characteristic response values ​​of each region are obtained to form the regional characteristic response distribution value.

[0122] The benefit of the formula is that it effectively improves the ability to distinguish regional feature responses by introducing the comprehensive quantification of gradient direction change value, spectrum balance and color concentration.

[0123] S203: Mapping the multi-region response value grid to the image space position according to the regional feature response distribution value, establishing a regional weight distribution matrix, and obtaining a regional feature response weight map.

[0124] The specific implementation process first defines a spatial grid within the original image space. For example, the image is divided into grid cells of size 10 pixels × 10 pixels. Based on the grid cells, the spatial position coordinates of each grid cell within the region are calculated, and the label number b of the region to which each grid cell belongs is determined. Assuming that the grid cell at spatial position (50, 100) corresponds to the label region 1, and the feature response value of region 1 is 1.703 as calculated above, the feature response value of the grid position (50, 100) is set to 1.703. Similarly, the feature response values ​​of other spatial positions are mapped one by one. Subsequently, the feature response weights within each grid cell are calculated by weighted averaging the feature response values ​​of adjacent spatial positions using a spatial smoothing function, such as a Gaussian kernel function (with a radius of 3 pixels), to generate a regional weight distribution matrix. Assuming that the characteristic response values ​​of the adjacent grid cells at the spatial position (50, 100) are 1.703, 1.5, and 1.2, respectively, the weighted average calculation is performed using weights (0.6, 0.3, and 0.1): 1.703×0.6+1.5×0.3+1.2×0.1=1.5918, and the regional characteristic response weight value at this position is 1.592. This method is used to perform weighted smooth mapping on the entire image, and finally the regional characteristic response weight map is obtained.

[0125] See also Figure 4 , the steps for obtaining the image enhancement residual map are as follows:

[0126] S301: Decomposing the input image based on the regional feature response weight map, extracting the edge-guided feature map, texture map, and brightness map, performing regional feature response weight comparison, extracting high-frequency regional features, and generating multi-scale regional response feature values;

[0127] First, the input image is scaled. In the scale decomposition process, taking three scales as an example, Gaussian filtering is performed on the input image in turn, where the Gaussian filter kernel parameters are set to standard deviations σ of 0.5, 1.0, and 2.0 respectively, thereby obtaining three image hierarchies of different scales. The above image hierarchies are used to extract edge-guided feature maps, texture maps, and brightness maps scale by scale. In the extraction process, taking scale 1.0 as an example, the Sobel operator is first called to perform convolution operation on the scale image to obtain the edge-guided feature map of the image, and then the local variance calculation is used to obtain the texture map. The brightness map is generated by the grayscale average value of the scale image and the grayscale difference of the neighborhood. The specific execution process is: calculate the absolute value of the difference between the grayscale value of each pixel and the grayscale value of the neighborhood (3×3 window) and average them to obtain the brightness map. At this time, the brightness map value Y is obtained. c Expressed as an integer value, for example, if the grayscale values ​​of a pixel area at a certain position are 121, 125, 123, 126, 130, 128, 127, 129, and 131, then the brightness mapping value Y of the area is cCalculated as:

[0128]

[0129] Subsequently, for the above-mentioned images of different scales, the edge, texture, and brightness mapping data of the corresponding scales are extracted through the above steps and stored in array form respectively. The response values ​​of the edge-guided feature map, texture map, and brightness map are then compared. Based on the local area (for example, a 5×5 neighborhood), the above three eigenvalues ​​are normalized respectively, and the regional feature response weight comparison is performed pixel by pixel. The difference between the above three normalized eigenvalues ​​is calculated, and the threshold is set to 0.1 (the threshold is set based on the fact that when the eigenvalue difference exceeds 10% of the average value, it is a high-frequency area). If the edge-guided eigenvalue is greater than the threshold and is significantly higher than the texture and brightness features, the area is marked as a high-frequency area and the eigenvalue of the area is retained. Otherwise, it is ignored. The high-frequency area feature data is obtained, and multi-scale regional response feature values ​​are generated in this way.

[0130] S302: Call the multi-scale regional response characteristic value, calculate the response difference value, construct the characteristic sensitivity coefficient, and adjust the residual response with the brightness map difference value. The formula is:

[0131]

[0132] Calculate and obtain the weight-adjusted residual structure;

[0133] Among them, W res Represents the weight-adjusted residual structure, ΔN c represents the intensity difference of the edge-guided feature map at the cth scale, ΔG c Represents the structural difference of the texture map at the cth scale, Y c Represents the grayscale distribution value of the brightness map at the cth scale, M c Indicates the characteristic sensitivity coefficient, max c Indicates the maximum value operation among all scales;

[0134] For each scale, calculate the edge-guided feature intensity difference ΔN c and texture mapping structure difference ΔG c , where ΔN c The specific calculation process is the absolute value of the difference between the edge features of adjacent pixels. For example, if the edge response values ​​of position (x, y) and the neighborhood (x+1, y) are 0.7 and 0.4 respectively, then ΔN is calculated. c =|0.7-0.4|=0.3; and the texture mapping structure difference ΔG c is the absolute value of the difference in texture mapping values ​​in adjacent areas. For example, if the texture values ​​at the same position (x, y) and the neighboring area (x+1, y) are 0.6 and 0.2 respectively, then ΔG is calculated.c =|0.6-0.2|=0.4.

[0135] Then use the brightness mapping value Y c (For calculation, see example in paragraph S301) and the calculated response difference value to construct the characteristic sensitivity coefficient M c , where M c The average value of the response difference in the local area is used as the sensitivity coefficient. For example, for a 5×5 area, if the average response difference in the area is 0.25, the characteristic sensitivity coefficient is set to M c =0.25, and further call the formula to calculate the residual response weight adjustment structure W res :

[0136]

[0137] Combine the above sample data into the calculation, taking scale 1 as an example:

[0138]

[0139] By calculating all scales and taking the maximum value as the final weight to adjust the residual structure, for example, the calculation results of scales 1, 2, and 3 are 0.0070, 0.0095, and 0.0062 respectively, the final W res The maximum value is 0.0095.

[0140] S303: Adjust the residual structure according to the weight, fuse the residual signals of all scale maps, construct the multi-scale structural relationship of image elements, and generate an image enhancement residual map.

[0141] Substitute the aforementioned parameter W res =0.0095, the residual structure is adjusted by weight, and the residual signals of all scale maps are fused scale by scale. The specific execution process includes: first, for each scale map, the residual structure is multiplied by the residual signal value of the original scale map pixel by pixel, for example, Figure 1 The original residual signal of a pixel is 0.4, which is calculated as 0.0095×0.4=0.0038 after fusion. Similar operations are performed scale by scale to obtain image data after residual adjustment at multiple scales. Finally, weighted fusion processing is performed pixel by pixel on the image data after residual adjustment at different scales. The weight of weighted fusion is based on the original feature sensitivity coefficient M at different scales. c The ratio is determined. For example, if the characteristic sensitivity coefficients corresponding to scales 1, 2, and 3 are 0.25, 0.30, and 0.20 respectively, then the scale Figure 1 、 2 The fusion weights of 3 are 33.3%, 40.0%, and 26.7% respectively. The calculation of the residual map after pixel fusion at a certain position is:

[0142] Fusion pixel value = 0.0038 × 33.3% + 0.0055 × 40.0% + 0.0040 × 26.7% ≈ 0.00447

[0143] This method is used to complete the weighted fusion of pixels at all positions, and finally construct a complete image enhancement residual map.

[0144] Table 3 Scale response characteristics and sensitivity coefficient calculation data table

[0145]

[0146] As shown in Table 3, the data in the table show the calculation results of regional response characteristics and sensitivity coefficients at three scales, where the maximum W res = 0.0095 is used for subsequent scale fusion processing.

[0147] See also Figure 5 , the specific steps for obtaining the multi-scale enhanced fusion image are:

[0148] S401: Based on the image enhancement residual map, call the scale map brightness mean and edge direction distribution map, calculate the brightness offset difference and the edge direction angle value, screen the area that meets the edge consistency threshold condition, and obtain the edge consistency area distribution coefficient;

[0149] First, call the brightness mean of the scalogram. In the specific execution process, assume that the scalogram is divided into multiple sub-regions. For each region, calculate the sum of the pixel brightness values ​​and divide it by the number of pixels in the region to obtain the regional brightness mean. For example, assume that there are 10 pixels in a scalogram region, and their brightness values ​​are:

[0150] 156, 160, 158, 157, 161, 155, 159, 162, 156, 160, then by summing 156 + 160 + 158 + 157 + 161 + 155 + 159 + 162 + 156 + 160 = 1584, and then dividing it by the total number of pixels 10, we can get the average brightness of the area. Similarly, the corresponding brightness mean data is calculated for other scale areas; then, the edge direction distribution map is called. In the specific implementation process, the directional gradient method is used to extract the image edge gradient vector pixel by pixel, and the gradient direction angle value at each pixel is calculated. For example, the horizontal and vertical gradient amplitudes G are calculated point by point through the Sobel operator. x ,G y , and then calculate for each pixel If the horizontal gradient value of a pixel is 15 and the vertical gradient value is 10, then the edge direction angle at the pixel is calculated as After that, the brightness offset difference is calculated. The specific execution process is to call the brightness value in the residual map pixel by pixel and perform the difference operation with the above brightness mean. For example, if the brightness value of a pixel is 164, the brightness offset difference is calculated as |164-158.4|=5.6; then calculate the correlation between the above brightness offset difference and the edge direction angle. Specifically, compare the angle between the brightness difference direction and the edge direction between the pixels, that is, calculate the absolute difference between the gradient direction of the brightness change at each pixel and the angle between the edge direction. If the brightness change gradient direction at a certain pixel is 30° and the edge direction angle is 33.69°, the angle difference is |30-33.69|=3.69°; next, for the calculated angle difference Perform region screening, that is, judge and screen pixel regions whose angle difference is less than the preset edge consistency threshold. The specific edge consistency threshold setting is obtained through the statistics of the previous image data. Assuming that the angle difference distribution of 1000 typical images is analyzed through experiments, it is determined that the angle difference in the range of 0° to 5° belongs to the area with high consistency. The threshold is selected as 5°, and the region that meets the conditions is determined to be the edge consistent region; further, for the edge consistent region, the number of pixels that meet the threshold is counted and divided by the total number of pixels to calculate the edge consistent region distribution coefficient. Assuming that the number of pixels that meet the threshold condition is 680 and the total number of pixels involved in the calculation is 1000, the calculated edge consistent region distribution coefficient is Thus, the edge consistent area distribution coefficient is obtained.

[0151] S402: Extract the texture contrast value of the matching area based on the edge consistent area distribution coefficient, and construct the brightness difference and texture contrast difference ratio using the formula:

[0152]

[0153] Calculate the ratio of overall brightness deviation to texture distinctness, reorder regional brightness according to texture features, and obtain a brightness callback reference sequence;

[0154] Among them, V represents the ratio of overall brightness deviation to texture distinctness, u represents the number of pixels involved in the calculation in the edge consistent area, and Q d Represents the brightness value of the d-th pixel in the edge consistent area, Indicates the mean brightness of the pixel in the scale map, X d Indicates the texture contrast value corresponding to the d-th pixel in the texture contrast map, max(X d ) represents the maximum value of the texture contrast value corresponding to all participating pixels in the current area, K d,r Represents the brightness difference between the d-th pixel and the adjacent r-th pixel;

[0155] Substitute the result value of the edge consistent area distribution coefficient of the above parameters into 0.68, and then call the texture contrast value of the matching area. In the specific execution process, the texture contrast value is calculated by calling the image gray level co-occurrence matrix (GLCM) to quantitatively analyze the grayscale difference of the area near each pixel. Taking a certain pixel as an example, the pixel grayscale values ​​in its local window (for example, a 3×3 neighborhood window) are The texture contrast value X of this pixel is calculated based on GLCM d is 7.2. Similarly, the texture contrast is calculated for all pixels point by point. Then, the ratio of the brightness difference to the texture contrast is calculated. The participating parameters in the formula are expressed as follows: u represents the total number of pixels in the edge consistent area, which is set to 680 in this example, Q d Indicates the brightness value of the d-th pixel. For example, the brightness value of the pixel is 164. is the mean brightness (158.4), then Then with its texture contrast value X d =7.2, we get 5.6+7.2=12.8. After calculating and accumulating the 680 pixels, we assume that the cumulative total is 8500. Then we calculate the maximum value of the texture contrast in the participating pixel area, for example, assuming it is 12.0. At the same time, we calculate the brightness difference value K between the pixels. d,r For example, if the brightness of two adjacent pixels is 164 and 160 respectively, then the brightness difference is |164-160|=4. The total of all adjacent pixel differences is calculated, assuming it is 3000, and is substituted into the formula for calculation:

[0156]

[0157] The results show that the ratio of overall brightness deviation to texture distinctness is 0.236. The ratios obtained in each region are reordered from large to small to obtain the brightness callback reference sequence of the corresponding region.

[0158] S403: Based on the brightness callback reference sequence, call the brightness information of the matching area, perform brightness mapping and segment adjustment, integrate texture features and edge direction features, and obtain a multi-scale enhanced fusion image.

[0159] Bring in the brightness callback reference sequence result of the aforementioned parameters, call the matching area brightness information, and perform brightness mapping and segmented adjustment on the brightness value pixel by pixel. In the specific execution process, set the mapping function to linear transformation per region. Assuming that the original brightness range of a region is [150,170], and the brightness range after adjustment is mapped to [155,175], then when the original brightness of a pixel is 164, its corresponding adjusted brightness is:

[0160]

[0161] The adjusted brightness is integrated with the original texture features and edge direction features, and the pixel-by-pixel fusion operation is performed through the multi-scale fusion technology. In the specific fusion process, it is assumed that a fusion weight is assigned to each scale image. The weight depends on the comprehensive index of regional texture sharpness and edge direction consistency. For example, if the regional texture sharpness index is 0.72 and the edge direction consistency coefficient is 0.68, the fusion weight is set to The weight is multiplied with the brightness-adjusted image at the corresponding scale, and then accumulated and fused scale by scale to form a multi-scale enhanced fused image.

[0162] See also Figure 6 , the specific steps for obtaining the regional enhanced image result map are:

[0163] S501: Based on the multi-scale enhanced fusion image, call the edge response difference map of the original image and the enhanced image to obtain the regional edge response intensity, compare the gradient response value of the difference map at the regional edge position with the set edge enhancement discrimination threshold, and determine whether the region belongs to the high-frequency change region, and generate the high-frequency response interval value;

[0164] The specific execution process is as follows: first, call the edge response values ​​of the corresponding positions of the original image and the enhanced image, subtract the edge response values ​​of the corresponding positions in the enhanced image from the edge response values ​​of the corresponding positions in the original image, and obtain the edge response difference map corresponding to each regional position. Taking an actual scene as an example, randomly select the position (120,80) in the monitored image. Assuming that the edge response value of the original image is 34 and the edge response value of the enhanced image is 50, the difference is 50-34=16, and the value of the difference map at this position is 16. Next, the gradient response value of the edge position of the region is calculated for the difference map using the numerical difference method. Taking the position (120,80) as an example, the response values ​​of the difference map are taken for its adjacent positions (119,80), (121,80), (120,79), and (120,81), and are set to 14, 18, 12, and 20, respectively. At this time, the gradient response value of the position is calculated using the finite difference method. Next, the gradient response value is compared with the set edge enhancement threshold. The edge enhancement threshold is set based on experimental analysis of edge characteristics. For example, in the experiment, the edge enhancement threshold is set to 7.5 based on the edge response gradient statistics of a large number of images. The gradient response value of 8.94 at the current area position (120, 80) is greater than 7.5. Therefore, the location is determined to belong to the high-frequency change area, and the corresponding difference map value 16 is recorded in the high-frequency response interval value. Similarly, the same method is used to traverse other locations in the image to obtain the high-frequency response interval value set of the entire image.

[0165] S502: Call the high-frequency response interval value, combine the texture consistency offset value and the brightness change rate, and use the formula for all areas in the high-frequency response interval:

[0166]

[0167] Calculate the enhancement amplitude index, filter the area greater than the edge enhancement discrimination threshold, and obtain the edge enhancement effective coefficient distribution;

[0168] in, Indicates the enhancement amplitude index of the coordinate (e, s) position, U e,s Indicates the texture consistency offset value of the position, L e,s Indicates the rate of change of brightness, ΔP e,s represents the response value of the edge response difference map, represents the enhancement value of the x-th layer fusion image at the position under the differential scale, and v represents the number of scales involved in the fusion image;

[0169] Call the high frequency response interval value obtained above and the texture consistency offset value U e,s and the brightness change rate L e,s , respectively, are obtained through image local feature analysis. The texture consistency offset value uses the local standard deviation algorithm to analyze the difference between local pixels and central pixels. For example, if the local standard deviation of the position (120, 80) is 3.6, then the texture consistency offset value is 3.6; the brightness change rate is calculated by the grayscale difference of the image pixels before and after enhancement. For example, if the grayscale value of the original image is 125 and the grayscale value of the enhanced image is 135, the brightness change rate is (135-125) / 125=0.08. At the same time, the fusion enhancement value under the differential scale is obtained. Assuming that a three-layer fusion scale is currently used, the specific fusion enhancement value is expressed as follows using an actual example:

[0170] Table 4 Differentiated scale fusion enhancement value table

[0171] Number of layers Fusion Enhancement Value 1 1.5 2 1.2 3 0.8

[0172] As shown in Table 4, the fusion enhancement values ​​of this area are 1.5, 1.2, and 0.8 respectively. e,s The corresponding position (120,80) is 16. Substitute the above values ​​into the formula:

[0173]

[0174] After the above process is performed for all high-frequency response interval positions, the enhancement amplitude index of each position is obtained Next, the enhancement amplitude index at each position is compared and screened one by one. Continuing with the aforementioned edge enhancement threshold of 7.5, the actual comparison is performed. If the enhancement amplitude index at position (120, 80) is 0.0693 < 7.5, then this position is not considered an effective edge enhancement position. If the enhancement amplitude index at another position is 8.2 > 7.5, then this position is considered an effective edge enhancement area. After this individual comparison and screening, a complete distribution map of effective edge enhancement coefficients is obtained.

[0175] S503: Calling the edge enhancement effective coefficient distribution, identifying the area below the edge enhancement discrimination threshold, restoring the image value to the original image value, and maintaining the enhancement process for the remaining areas to obtain the regional enhanced image result map.

[0176] The edge enhancement coefficient distribution map obtained above is used to analyze each region one by one. For example, the enhancement amplitude index corresponding to position (120, 80) is 0.0693, which is determined to be lower than the edge enhancement threshold of 7.5. Therefore, the image value at that position is restored to the original image value of 125. If the enhancement amplitude index at another position (150, 90) is 8.2>7.5, then the enhanced value of that position is retained at 140. After performing the above operation on all positions, the complete region-enhanced image result is obtained.

[0177] The image enhancement system based on deep learning is used to execute the above-mentioned image enhancement method based on deep learning. The system includes:

[0178] The image frequency partition module obtains the input image pixel matrix, extracts the brightness difference, gradient direction change value and local change trend intensity, makes judgments based on the product of the brightness difference and texture density distribution and the gradient direction change value, divides the frequency region and marks the image block type, and generates the image frequency partition label;

[0179] The regional feature mapping module calls the image frequency partition label to extract the regional edge intensity value, texture fineness value and contrast value, normalizes them and assigns them to the corresponding image block position with weighted superposition to generate the regional feature response weight map;

[0180] The scale residual fusion module calls the regional feature response weight map to obtain the image block content and extract the multi-scale edge-guided map, texture map and brightness map. It performs residual calculation on the three maps according to the image block weight value and performs weighted fusion. The fusion value is superimposed according to the image block position to reconstruct the entire image and generate an image enhancement residual map.

[0181] The edge brightness callback module calls the image enhancement residual map, collects the brightness mean, edge direction distribution map and texture contrast map, determines the regional matching relationship based on the edge direction consistency coefficient and performs brightness linear callback to generate a multi-scale enhanced fusion image;

[0182] The high-frequency detail restoration module calls the multi-scale enhanced fusion image, extracts the edge response difference map between it and the original image, and combines the texture consistency offset value and the brightness change rate for combined judgment. The unsatisfactory area is restored to the original image value to generate the regional enhanced image result map.

[0183] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An image enhancement method based on deep learning, characterized in that: The following steps are involved: S1: Obtain the input image pixel matrix, determine the regional frequency attribute based on the brightness difference, gradient direction change value, local change trend intensity and texture density distribution, divide it into high-frequency area and low-frequency area according to the edge response threshold, and obtain the image frequency partition label; The steps for obtaining the image frequency partition label are specifically as follows: S101: Based on the input image pixel matrix, calculate the grayscale difference and compare it with the brightness difference threshold, count and normalize the number of pixels with significant brightness difference, and obtain the brightness difference value; S102: Call the brightness difference value, and use the formula: Obtain the regional gradient response value and superimpose it with the brightness difference distribution to obtain the frequency mutation factor value; Among them, F Δg Represents the regional gradient response value, Δg a Represents the gradient direction change value of the a-th pixel pair, G a Indicates the local change trend strength of the point, H a is the point gradient change density value, B a is the brightness difference of the point neighborhood, and N represents the number of pixel pairs participating in the statistics; S103: comparing the frequency mutation factor value with the texture density distribution and the edge response threshold, determining the regional frequency attribute, and generating an image frequency partition label; S2: Based on the image frequency partition label, extract the edge intensity value, texture fineness value, contrast value, gradient direction change value, spectrum balance value and color distribution concentration of multiple regions, perform nonlinear cross integration operation, assign the results to the corresponding region, and obtain the regional feature response weight map; The steps for obtaining the regional feature response weight map are specifically as follows: S201: extracting edge intensity values, texture fineness values, and contrast values ​​of multiple regions based on the image frequency partition labels, normalizing them according to the maximum and minimum values, and generating multi-region standardized image feature values; S202: Calling the multi-region standardized image feature values, collecting gradient direction change values, spectrum balance values ​​and color distribution concentration, performing nonlinear cross integration operations, using the formula: Calculate and obtain regional characteristic response values, adjust the attribute response ratio within the region, and generate regional characteristic response distribution values; Among them, RC b represents the characteristic response value of region b, R b represents the normalized edge strength value of the bth region, F b represents the normalized texture fineness value of the b-th region, O b represents the normalized contrast value of the b-th region, G b Indicates the gradient direction change value of the b-th region, W b Indicates the spectrum balance value of the b-th area, Z b represents the color distribution concentration of the b-th area, ∑(W b +Z b ) represents the joint summation of spectrum and color; S203: Mapping the multi-region response value grid to the image space position according to the regional feature response distribution value, establishing a regional weight distribution matrix, and obtaining a regional feature response weight map; S3: Decomposing the input image according to the regional feature response weight map, extracting a multi-scale edge-guided feature map, a texture map, and a brightness map, adjusting the residual response values ​​of the multiple maps according to the weights, enhancing the details of the high-frequency region and suppressing the redundancy of the low-frequency region, fusing all the scale maps and constructing a mapping structure to obtain an image enhancement residual map; The steps for obtaining the image enhancement residual map are specifically as follows: S301: Decomposing the input image on a scale according to the regional feature response weight map, extracting the edge-guided feature map, the texture map, and the brightness map, performing regional feature response weight comparison, extracting high-frequency regional features, and generating multi-scale regional response feature values; S302: Call the multi-scale regional response characteristic value, calculate the response difference value, construct the characteristic sensitivity coefficient, and adjust the residual response with the brightness map difference value, using the formula: Calculate and obtain the weight-adjusted residual structure; Among them, W res Represents the weight-adjusted residual structure, ΔN c represents the intensity difference of the edge-guided feature map at the cth scale, ΔG c Represents the structural difference of the texture map at the cth scale, Y c Represents the grayscale distribution value of the brightness map at the cth scale, M c Indicates the characteristic sensitivity coefficient, max c Indicates the maximum value operation among all scales; S303: adjusting the residual structure amount according to the weight, fusing the residual signals of all scale maps, constructing the multi-scale structural relationship of image elements, and generating an image enhancement residual map; S4: Based on the image enhancement residual map, call the scale map brightness mean, edge direction distribution map and texture contrast map, match the area according to the edge consistency coefficient and perform brightness callback to obtain a multi-scale enhanced fusion image; The steps for acquiring the multi-scale enhanced fusion image are specifically as follows: S401: Based on the image enhancement residual map, call the scalogram brightness mean and edge direction distribution map, calculate the brightness offset difference and the edge direction angle value, screen the area that meets the edge consistency threshold condition, and obtain the edge consistency area distribution coefficient; S402: Extract the texture contrast value of the matching area based on the edge consistent area distribution coefficient, and construct the brightness difference and texture contrast difference ratio using the formula: Calculate the ratio of overall brightness deviation to texture distinctness, reorder regional brightness according to texture features, and obtain a brightness callback reference sequence; Among them, V represents the ratio of overall brightness deviation to texture distinctness, u represents the number of pixels involved in the calculation in the edge consistent area, and Q d Represents the brightness value of the d-th pixel in the edge consistent area, Indicates the mean brightness of the pixel in the scale map, X d Indicates the texture contrast value corresponding to the d-th pixel in the texture contrast map, max(X d ) represents the maximum value of the texture contrast value corresponding to all participating pixels in the current area, K d,r Represents the brightness difference between the d-th pixel and the adjacent r-th pixel; S403: Based on the brightness callback reference sequence, call the matching area brightness information, perform brightness mapping and segment adjustment, integrate texture features and edge direction features, and obtain a multi-scale enhanced fusion image.

2. The image enhancement method based on deep learning according to claim 1, characterized in that The image frequency partition labels are specifically high-frequency areas and low-frequency areas. The regional feature response weight map includes multi-region edge intensity values, texture fineness values, and contrast values. The image enhancement residual map is specifically a multi-scale edge-guided feature map, texture map, and brightness map after scale decomposition. The multi-scale enhanced fusion image includes the scale map brightness mean, edge direction distribution map, and texture contrast map.

3. The image enhancement method based on deep learning according to claim 2, characterized in that: The method further comprises: S5: Based on the multi-scale enhanced fusion image, the edge response difference map, texture consistency offset value and brightness change rate of the original image and the enhanced image are called, and an enhancement amplitude judgment operation is performed on all high-frequency areas in the image. The area with a response amplitude lower than the edge enhancement judgment threshold is eliminated, and the original image value is restored to the area, and a regional enhanced image result map is output; The region-enhanced image result diagram is specifically an image after the high-frequency region enhancement amplitude is determined.

4. The image enhancement method based on deep learning according to claim 3, characterized in that The steps for obtaining the region enhanced image result map are specifically as follows: S501: Based on the multi-scale enhanced fusion image, call the edge response difference map of the original image and the enhanced image to obtain the regional edge response strength, compare the gradient response value of the difference map at the regional edge position with the set edge enhancement discrimination threshold, and determine whether the region belongs to the high-frequency change region, and generate a high-frequency response interval value; S502: Call the high frequency response interval value, combine the texture consistency offset value and the brightness change rate, and use the formula for all areas in the high frequency response interval: Calculate the enhancement amplitude index, filter the area greater than the edge enhancement discrimination threshold, and obtain the edge enhancement effective coefficient distribution; in, Indicates the enhancement amplitude index of the coordinate (e, s) position, U e,s Indicates the texture consistency offset value of the position, L e,s Indicates the rate of change of brightness, ΔP e,s represents the response value of the edge response difference map, represents the enhancement value of the x-th layer fusion image at the position under the differential scale, and v represents the number of scales involved in the fusion image; S503: calling the edge enhancement effective coefficient distribution, identifying the area below the edge enhancement discrimination threshold, restoring the image value to the original image value, and maintaining the enhancement process for the remaining areas to obtain the regional enhanced image result map.

5. Image enhancement system based on deep learning, characterized by: The deep learning-based image enhancement method according to any one of claims 1 to 4, wherein the system comprises: The image frequency partition module obtains the input image pixel matrix, extracts the brightness difference, gradient direction change value and local change trend intensity, makes judgments based on the product of the brightness difference and texture density distribution and the gradient direction change value, divides the frequency region and marks the image block type, and generates the image frequency partition label; The regional feature mapping module calls the image frequency partition label, extracts the regional edge intensity value, texture fineness value and contrast value, normalizes them and weightedly superimposes them and assigns them to the corresponding image block position to generate a regional feature response weight map; The scale residual fusion module calls the regional feature response weight map, obtains the image block content and extracts the multi-scale edge-guided map, texture map and brightness map, performs residual calculation on the three maps according to the image block weight value and performs weighted fusion, and superimposes the fusion value according to the image block position to reconstruct the entire image and generate an image enhancement residual map; The edge brightness callback module calls the image enhancement residual map, collects the brightness mean, edge direction distribution map and texture contrast map, determines the regional matching relationship based on the edge direction consistency coefficient and performs brightness linear callback to generate a multi-scale enhanced fusion image; The high-frequency detail restoration module calls the multi-scale enhanced fusion image, extracts the edge response difference map between it and the original image, and makes a combined judgment based on the texture consistency offset value and the brightness change rate. It determines that the unsatisfactory area is restored to the original image value and generates a regional enhanced image result map.

Images