Image adaptive enhancement method and system based on deep learning

Abstract

The invention discloses an image adaptive enhancement method and system based on deep learning, and relates to the field of image processing, and the system comprises an image processing platform which is provided with an image collection module, an image processing module, an image analysis module, a model management module and an image management module; the image acquisition module is used for inputting image data and image information; the image processing module is used for acquiring segmented image areas and image quality data of corresponding pixel points; the image analysis module is used for acquiring a comprehensive classification result of pixel points corresponding to the segmented image region; the model management module is used for constructing an image processing link and a corresponding image processing link model; the image management module is used for acquiring a region enhancement image corresponding to the segmented image region according to an image processing link and a corresponding image processing link model, and acquiring a self-adaptive enhancement image of the image data; according to the invention, the accuracy in the image enhancement process is improved to a certain extent.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method and system for adaptive image enhancement based on deep learning. Background Art

[0002] In today's digital age, images, as an important information carrier, are widely used in various fields. However, in the process of image acquisition, transmission and storage, they are often affected by various factors, resulting in a decrease in image quality. Traditional image enhancement methods mainly include improving image contrast and clarity based on histogram equalization, filtering and other technologies, but there are still some limitations. For example, histogram equalization can lead to over-enhancement of the image and loss of image detail information, and filtering methods may blur the image edges, thereby affecting the image resolution.

[0003] An image adaptive enhancement algorithm for real-time image enhancement with publication number CN112991240B discloses an image adaptive enhancement algorithm for real-time image enhancement, including: obtaining a grayscale image to be processed in the current frame; automatically calculating the effective grayscale threshold of the grayscale histogram corresponding to the grayscale image based on an iterative algorithm; performing a grayscale equal-pitch density equalization transformation on the grayscale histogram according to the effective grayscale threshold; during the transformation process, when calculating the mapped grayscale value corresponding to the enhanced image, a threshold limit is imposed on the spacing after the grayscale is equally spaced and rearranged, and a grayscale bias correction is performed on the mapped grayscale value; after the grayscale equal-pitch density equalization transformation is completed, an enhanced image corresponding to the grayscale image is obtained.

[0004] However, the image enhancement process achieved by merely analyzing and processing the grayscale value of the image cannot be personalized according to the user's needs for the corresponding image, and the accuracy of the image enhancement process is insufficient. With the rapid development of deep learning technology, it has achieved great success in image recognition, image processing and other fields, which also provides new ideas for image adaptive enhancement.

[0005] Therefore, how to adaptively process various parts of the image based on deep learning methods, avoid over-enhancement of different positions within the image, and improve the accuracy and personalization of the image enhancement process is a problem we need to solve. To this end, an image adaptive enhancement method and system based on deep learning are provided. Summary of the invention

[0006] In order to solve the above technical problems, the purpose of the present invention is to provide an image adaptive enhancement method and system based on deep learning.

[0007] The object of the present invention can be achieved by the following technical solutions: an image adaptive enhancement system based on deep learning includes an image processing platform, wherein the image processing platform is provided with an image acquisition module, an image processing module, an image analysis module, a model management module and an image management module;

[0008] The image acquisition module is used to input image data and corresponding image information, generate image data packets and send them to the image processing module;

[0009] The image processing module is used to perform region classification on the corresponding image according to the image data and the image information, perform image segmentation according to the region classification result, and obtain image quality data of the segmented image region and the corresponding pixel points;

[0010] The image analysis module is used to perform classification processing according to the image quality data corresponding to the pixel points in the segmented image area, and obtain the comprehensive classification results of the pixel points corresponding to the segmented image area;

[0011] The model management module is used to set image processing sub-models according to different comprehensive classification results, comprehensively sort the image processing sub-models, and construct image processing links and corresponding image processing link models;

[0012] The image management module is used to input the obtained segmented image regions into the image processing link respectively, obtain the regional enhanced images of the corresponding segmented image regions according to the image processing link model, merge the regional enhanced images, and obtain the adaptive enhanced image corresponding to the image data.

[0013] Furthermore, the process of the image acquisition module inputting image data and corresponding image information includes:

[0014] Set up image input terminal and information input terminal;

[0015] The image input terminal is used to input image data and mark the image data; the information input terminal is used to input image information corresponding to the image data, and the image information includes image description information and image key information;

[0016] The obtained image data and image information are used to generate an image data packet according to the marking processing result.

[0017] Furthermore, the process of the image processing module acquiring the image quality data of the segmented image area and the corresponding pixel points includes:

[0018] Obtaining pixel values ​​corresponding to corresponding pixel points according to the image data, obtaining a grayscale image corresponding to the image data according to the pixel values, performing edge detection processing on the grayscale image to determine corresponding edge pixel points, and generating an edge image according to position marks of the edge pixel points in the grayscale image;

[0019] Setting a semantic analysis algorithm, analyzing and processing the image information through the semantic analysis algorithm, extracting corresponding descriptive semantic information and feature semantic information, comparing and analyzing the descriptive semantic information and feature semantic information with corresponding pixel points in the image data, determining corresponding pixel point positions, and marking the descriptive semantic information and feature semantic information according to the pixel point positions;

[0020] According to the distribution of the corresponding edge pixel points in the edge image, the image sub-region is set, and according to the distribution results of the description semantic information and the feature semantic information in each image sub-region, it is determined whether to segment and extract the corresponding image sub-region, and the segmented image region after segmentation and extraction is obtained;

[0021] Mark the segmented image area as the original image, set a mapping function, scale the pixel values ​​of the corresponding pixel points of the original image based on the mapping function, and obtain the corresponding scaled image;

[0022] According to the mean square error of pixel values ​​between the original image and the scaled image, it is marked as image quality data.

[0023] Furthermore, the process of the image analysis module obtaining the comprehensive classification result of the pixel points corresponding to the segmented image area includes:

[0024] Obtain the distribution of pixel points corresponding to the description semantic information and the feature semantic information in the segmented image area, and divide the corresponding pixel points into key pixel points, basic pixel points and auxiliary pixel points according to the marking results of the corresponding pixel points;

[0025] Acquire image quality data and historical image data packets within the segmented image area, preset pixel quality level intervals according to the historical image data packets, namely, high-quality pixel intervals, medium-quality pixel intervals, and low-quality pixel intervals, compare and analyze the image quality data with the corresponding pixel quality level intervals, and determine the pixel quality level to which the pixel point corresponding to the corresponding image quality data belongs according to the comparison and analysis results;

[0026] The classification results corresponding to the above two different classification methods are obtained, and the obtained classification results are marked as the comprehensive classification results of the corresponding pixel points.

[0027] Furthermore, the process of the model management module setting the image processing sub-model according to different comprehensive classification results includes:

[0028] Obtain the image processing process of the image data pixels corresponding to different comprehensive classification results, and generate image processing data sets corresponding to the corresponding types of comprehensive classification results;

[0029] Based on the deep learning algorithm, the obtained image processing data set is analyzed and trained to construct an image processing sub-model corresponding to the corresponding type of comprehensive classification results.

[0030] Furthermore, the process of the model management module constructing the image processing link and the corresponding image processing link model includes:

[0031] Obtaining the importance level of the pixel points corresponding to each comprehensive classification result in the image processing process, sorting the corresponding comprehensive classification results according to the importance level, setting the corresponding image processing space in sequence according to the sorting results, and correlating the image processing space with the corresponding image processing sub-model, inputting the pixel point area corresponding to the corresponding comprehensive classification result into the image processing space, connecting the various image processing spaces in series, and generating an image processing chain;

[0032] The image merging process of each image processing space in the image processing link is obtained to generate a link data set. The constructed link data set is analyzed and processed based on the deep learning algorithm to construct an image processing link model.

[0033] Furthermore, the process of the image management module acquiring the adaptive enhanced image corresponding to the image data includes:

[0034] Obtain the comprehensive classification results of the corresponding pixel point areas in the segmented image area, input them into the idle image processing chain, and the corresponding image processing sub-models in each image processing space in the image processing chain analyze and process the corresponding pixel point areas in turn until the enhanced analysis of the corresponding segmented image area is completed in the image processing chain, and then input them into the image processing chain model for analysis and processing, and the corresponding segmented image area is enhanced.

[0035] The edge pixel points corresponding to the regional enhanced images are obtained, and the regional enhanced images are merged according to the distribution of the edge pixel points of each regional enhanced image to obtain an adaptive enhanced image corresponding to the corresponding image data.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. By setting image information, the key information of the image that people subjectively consider is marked, thereby improving people's personalized intervention in the image processing process. In addition, the classification results of different pixels in the image are obtained through image information and image data, and targeted enhancement processing is performed according to the classification results, which improves the accuracy of the image data processing process to a certain extent;

[0038] 2. Set the corresponding segmented image area according to the classification result of the image data, distribute the image information in each segmented image area, and perform image enhancement processing on each segmented image area, which avoids the influence of image processing failure or processing error on the entire image data to a certain extent;

[0039] 3. By setting up an image processing link, the pixel points corresponding to different classification results in each segmented image area are classified and processed, which ensures that each pixel point can be enhanced and improves the accuracy of the image enhancement process to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a schematic diagram of the image adaptive enhancement system based on deep learning in this application embodiment.

[0041] Figure 2 This is a schematic diagram of the image adaptive enhancement method based on deep learning according to an embodiment of the present application. DETAILED DESCRIPTION

[0042] like Figure 1 As shown, the image adaptive enhancement system based on deep learning includes an image processing platform, in which an image acquisition module, an image processing module, an image analysis module, a model management module and an image management module are arranged.

[0043] The image processing platform is used to acquire image data that needs to be analyzed and processed, and analyze the acquired image data based on a deep learning algorithm to achieve adaptive enhancement of the corresponding image data, thereby improving image quality, and store its processing process. The deep learning algorithm is updated according to the processing process, thereby better preparing for image data processing.

[0044] The image acquisition module is provided with an image input terminal and an information input terminal;

[0045] The image input terminal is used to input image data, mark the obtained image data, and store the image data after the marking process;

[0046] The information input terminal is used to input corresponding image information based on the marking result of the input image data, and the image information includes image description information and image key information, wherein:

[0047] The image description information is textual description information corresponding to the corresponding image data;

[0048] Image key information is key feature information corresponding to the corresponding image data;

[0049] It should be further explained that, in the specific implementation process, the image acquisition module sets the corresponding image data packet with the marking result corresponding to the obtained image data, and performs data association with other image information corresponding to the corresponding image data, stores the image data and image information in the corresponding image data packet, and processes and marks the obtained image data packet, and stores them in the image processing platform in three states: pending, processing and completed.

[0050] The image processing module is used to classify the corresponding image regions according to the image data and the image information, perform image segmentation according to the regional classification results, and obtain image quality data of the segmented image regions and corresponding pixels. The specific implementation process includes:

[0051] Setting an image recognition unit and an image preprocessing unit;

[0052] The image recognition unit obtains the image data and image information included in the image data packet marked as to be processed, and performs image recognition processing on the obtained image data and image information:

[0053] Obtain pixel value data corresponding to the pixel point of the corresponding image data, perform weighted average calculation on the obtained pixel value data based on the RGB three channels, and obtain the grayscale value of the corresponding pixel point;

[0054] The gray value of the corresponding pixel is marked as gray, and the values ​​of each channel corresponding to each pixel are marked as G, R and B respectively, where:

[0055] gray=0.299×R+0.587×G+0.114×B;

[0056] Performing statistical processing on the grayscale values ​​corresponding to each pixel of the image data to obtain a grayscale image corresponding to the image data;

[0057] Acquire image information in an image data packet, set a corresponding semantic analysis algorithm, perform semantic analysis on the corresponding image information based on the semantic analysis algorithm, and extract corresponding descriptive semantic information and feature semantic information;

[0058] Perform semantic repetition screening and analysis based on the obtained descriptive semantic information and characteristic semantic information, and mark the repeated semantic information as characteristic semantic information;

[0059] Setting image processing nodes according to the description semantic information, setting key image processing nodes according to the feature semantic information, and marking the image processing nodes and key image processing nodes at the positions of corresponding pixels of the image data based on a deep learning algorithm;

[0060] Obtain the grayscale values ​​of each adjacent pixel in the grayscale image and perform edge detection on the grayscale values ​​in the grayscale image. The specific implementation process includes:

[0061] The corresponding pixel coordinates in the grayscale image are marked as i and j, respectively, where i is the horizontal coordinate in the grayscale image, j is the vertical coordinate in the grayscale image, i = 0, 1, 2, ..., n-1 and n; j = 0, 1, 2, ..., m-1 and m; X and Y are respectively recorded as the horizontal direction and vertical direction of the grayscale image;

[0062] Setting an edge detection gradient template, wherein the edge detection gradient template is a corresponding 2×2, wherein 2 is the detection range corresponding to the edge detection gradient template;

[0063] Obtain the corresponding grayscale value in the edge detection template, and perform finite difference calculation on the corresponding grayscale value in the template in the horizontal and vertical directions to obtain the corresponding gradient images in the two directions. The specific implementation process includes:

[0064] G x (i,j)=I(i,j)-I(i+1,j+1);

[0065] G y (i,j)=I(i+1,j)-I(i,j+1);

[0066] Get the gradient amplitude of the pixel point in the corresponding edge detection gradient template in the corresponding grayscale image, and mark the gradient amplitude as TF, where TF = |G x |+|G y |;

[0067] A gradient threshold is preset and marked as TY, and the obtained gradient amplitude is compared and analyzed with the preset gradient threshold:

[0068] When TF>TY, the pixel is an edge pixel;

[0069] When TF≤TY, the pixel is not an edge pixel;

[0070] Mark the edge points in the grayscale image, and generate an edge image according to the marking results;

[0071] The image processing nodes and key image processing nodes in the image data are mapped to the edge image respectively, and image recognition and division are performed according to the mapping results. The specific implementation process includes:

[0072] Obtain the corresponding edge pixel points in the edge image, set the image sub-region according to the distribution of the edge pixel points, mark the corresponding image sub-region, obtain the marking symbol of the image sub-region, mark the corresponding image processing nodes and the number of key image processing nodes in the corresponding image sub-region according to the marking symbol, and obtain the corresponding number marking result;

[0073] The weight coefficients of the image processing nodes and key image processing nodes are preset, and the corresponding image sub-regions are analyzed and processed according to the weight coefficients and the quantity marking results. The specific implementation process includes:

[0074] The number of image processing nodes and key image processing nodes are marked as TS and GS respectively, the weight coefficients are marked as α and β respectively, and the comprehensive weight data in the corresponding image sub-region is marked as ZQ, where:

[0075] ZQ = TS × α + GS × β;

[0076] The average weight data and the fluctuation coefficient in the preset edge image are obtained, and the corresponding weight standard interval is obtained according to the product of the average weight data and the fluctuation coefficient;

[0077] Compare and analyze the comprehensive weight data and weight standard interval corresponding to each image sub-area:

[0078] If the comprehensive weight data belongs to the weight standard interval, the corresponding image sub-region is segmented and extracted, and marked as a segmented image region;

[0079] If the comprehensive weight data does not belong to the weight standard interval, the corresponding image sub-region is not segmented and extracted, and the image sub-region is divided or merged according to the relationship between the comprehensive weight data and the weight standard interval, and segmentation and extraction are performed according to the division or merging result to obtain the corresponding segmented image region;

[0080] If the comprehensive weight data is less than the weight standard interval, the adjacent image sub-region is merged with the image sub-region, and the merged image sub-region is segmented and extracted to obtain the corresponding segmented image region;

[0081] If the comprehensive weight data is greater than the weight standard interval, the image sub-region is divided, and the divided image sub-region is segmented and extracted to obtain the corresponding segmented image region.

[0082] The image preprocessing unit is used to obtain the segmented image area corresponding to the corresponding image data, and analyze and process the segmented image area to determine the image quality data of the corresponding segmented image area. The specific implementation process includes:

[0083] Obtain the pixel value of the corresponding pixel point in the corresponding segmented image area, and scale the corresponding segmented image area according to the pixel value corresponding to the pixel point, wherein:

[0084] Set the mapping function to mark the segmented image area as the original image, the coordinates of its pixel points are (x, y), and the scaled coordinates are (x s ,y s ), get the scaled coordinates (x s ,y s ) in the original image. j ,y j );

[0085] Get the actual position coordinates (x j ,y j ) around the four pixel points (x0, y0), (x0, y1), (x1, y0) and (x1, y1);

[0086] The values ​​of the scaled pixels are obtained based on the bilinear interpolation formula, and the four pixels are respectively denoted as S 11 , S 12 , S 21 and S 22 ; Mark the pixel value of the interpolated pixel as P, where:

[0087] P=(x1-x s )(y1-y s )S 11 +(x s -x0)(y1-y s )S 21 +(x1-x s )(y2-

[0088] y s )S 12+ (x s -x1)(y2-y s )S 22 ;

[0089] According to the pixel values ​​of the corresponding pixels in the corresponding segmented image area after scaling;

[0090] Get the number of rows and columns of image pixels in the segmented image area, and record them as m and n respectively; mark the pixel value of the original image as I(i, j), and the pixel value of the segmented image area after scaling as S(i, j); get the mean square error of the pixel values ​​between the original image and the scaled image, and mark it as MSE, where:

[0091]

[0092] The image quality data of the pixel points corresponding to the segmented image area is set according to the obtained mean square error value. The larger the image quality data is, the worse the image quality is.

[0093] It should be further explained that, in the specific implementation process, the image processing module stores the data information obtained in the image recognition unit and the image preprocessing unit into an image data packet corresponding to the image data, and sends the image data packet to the image analysis module.

[0094] The image analysis module is used to obtain the image data packet sent by the image processing module, analyze and process the image data according to the content in the image data packet, and classify the image data;

[0095] Set up type analysis unit and comprehensive analysis unit;

[0096] The type analysis unit is used to obtain image quality data and image description data in the image data packet, and perform image analysis on the image data according to the data type;

[0097] Get the image processing nodes and key image processing nodes in the segmented image area corresponding to the image data, and mark the corresponding pixel positions in the image processing nodes and key image processing nodes:

[0098] When the corresponding pixel point in the segmented image area corresponds to a key image processing node, it is marked as a key pixel point;

[0099] When the corresponding pixel point in the segmented image area corresponds to an image processing node, it is marked as a basic pixel point;

[0100] When the corresponding pixel points in the segmented image area do not correspond to the corresponding key image processing nodes and image processing nodes, they are marked as auxiliary pixel points;

[0101] Obtain image quality data of corresponding pixels in the segmented image area corresponding to the image data, analyze and process the image quality data, and classify the corresponding pixels;

[0102] Obtain a historical image data packet marked as processed in the image processing platform, and set pixel quality level intervals according to image quality data in the historical image data packet, which are respectively set to a high-quality pixel interval, a medium-quality pixel interval, and a low-quality pixel interval;

[0103] The image quality data corresponding to the high-quality pixel interval is smaller than the image quality data corresponding to the medium-quality pixel interval, and the image quality data corresponding to the medium-quality pixel interval is smaller than the image quality data corresponding to the low-quality pixel interval;

[0104] Compare and analyze the image quality data corresponding to the segmented image area with the corresponding pixel quality level intervals, and determine the pixel quality level to which the pixel point corresponding to the corresponding image quality data belongs according to the comparison results;

[0105] The comprehensive analysis unit obtains the pixel point types to which the corresponding pixel points obtained by the type analysis unit belong respectively;

[0106] Obtaining classification types corresponding to pixel points, recording key pixel points, basic pixel points and basic pixel points as one group of classification types, and recording high-quality pixel levels, medium-quality pixel levels and low-quality pixel levels as another group of classification types;

[0107] Perform cross-combination from two classification types, and set corresponding comprehensive classification results according to the cross-combination classification results;

[0108] The classification results of the corresponding pixel points are matched with the comprehensive classification results respectively, the classification type of the corresponding pixel points is analyzed according to the matching results, and uploaded to the image data packet, and the obtained image data packet is sent to the model management module.

[0109] The model management module is used to obtain the corresponding image data packet and set the image processing model according to the corresponding image data packet. The specific implementation process includes:

[0110] The model management module is equipped with a model training unit and a model management unit.

[0111] The model training unit is used to obtain image data packets marked as processed, classify the processed image data packets, extract the corresponding type of pixel enhancement process, and set the image enhancement process of the same type of pixels as an image processing data set, and the image processing data sets are respectively corresponding training sets and verification sets;

[0112] Based on the deep learning algorithm, the obtained training set is analyzed and trained to obtain the image processing sub-model corresponding to the corresponding classification type, until the corresponding loss function in the image processing sub-model tends to be stable, and the corresponding image processing sub-model is output, and the obtained verification set is input into the corresponding image processing sub-model for verification analysis. According to the verification analysis results, it is determined whether the corresponding model needs to be repeatedly trained. If it meets the verification standard, the corresponding image processing sub-model is output;

[0113] The model management unit obtains image processing sub-models corresponding to different types of comprehensive classification results, and sets an image processing link model according to the obtained image processing sub-models. The specific implementation process includes:

[0114] Analyze and process the link data set, set importance levels for the corresponding comprehensive classification results, and sort the corresponding comprehensive classification results according to the importance levels;

[0115] Among them, importance weight factors corresponding to key pixels, basic pixels, basic pixels, high-quality pixel levels, medium-quality pixel levels, and low-quality pixel levels are set respectively;

[0116] Obtain the importance weight factor of the corresponding classification result in the corresponding comprehensive classification result according to the cross-combination result, and obtain the importance level data of the corresponding comprehensive classification result according to the obtained importance weight factor;

[0117] Sort the data from low to high according to the importance level;

[0118] According to the sorting results, the corresponding image processing space is set in sequence, the image processing sub-model corresponding to the comprehensive classification result is input into the corresponding image processing space, the pixel point area corresponding to the corresponding comprehensive classification result is input into the image processing space, and the image processing sub-model analyzes and processes the corresponding pixel point area in the corresponding image processing space. After the processing is completed, it is input into the next image processing space according to the sorting result;

[0119] The image merging process of each image processing space in the image processing link is obtained to generate a link data set. The constructed link data set is analyzed and processed based on the deep learning algorithm to construct an image processing link model.

[0120] The image management module is used to obtain the corresponding image processing link model, analyze and process the image data packets to be processed, and complete the image adaptive enhancement process. The specific implementation process includes:

[0121] The image management module copies the image processing link set by the model management module and sets multiple image processing links. The multiple image processing links have the same functions, which makes it easier for the image processing platform to process multiple image data at the same time and detect the working status of the corresponding image processing links.

[0122] Obtain the comprehensive classification results of the corresponding pixel point areas in the segmented image area, input them into the idle image processing chain, and the corresponding image processing sub-models in each image processing space in the image processing chain analyze and process the corresponding pixel point areas in turn until the enhanced analysis of the corresponding segmented image area is completed in the image processing chain, and then input them into the image processing chain model for analysis and processing, and the corresponding segmented image area is enhanced.

[0123] Obtain edge pixel points corresponding to the regional enhanced image, merge the regional enhanced images according to the distribution of edge pixel points of each regional enhanced image, and obtain an adaptive enhanced image corresponding to the corresponding image data. The specific implementation process includes:

[0124] Obtain the edge pixel points corresponding to the regional enhanced image, perform weighted processing on the obtained pixel points, obtain the weight data of the corresponding edge detection result, and mark it as w n (x), mark the pixel value of the nth input image as v n (x), and mark the fused pixel value as z fused (x), where:

[0125]

[0126] In the fusion process, pixels are weighted according to the edge detection results to ensure that the pixel values ​​of the edge areas in the regional enhanced image remain unchanged or transition smoothly, thereby realizing the merging process of the regional enhanced images.

[0127] like Figure 2 As shown, the image adaptive enhancement method based on deep learning includes the following steps:

[0128] Step 1: Setting up an image processing platform, inputting image data and corresponding image information, and generating an image data packet according to the input image data and image information;

[0129] Step 2: Perform regional classification on the corresponding image according to the image data and image information respectively, perform image segmentation according to the regional classification result, and obtain image quality data of the segmented image area and the corresponding pixel points;

[0130] Step 3: Perform classification processing based on the image quality data of the corresponding pixels in the segmented image area to obtain the comprehensive classification results of the corresponding pixels in the segmented image area;

[0131] Step 4: Set up image processing sub-models according to different comprehensive classification results, comprehensively sort the image processing sub-models, and construct image processing links and corresponding image processing link models;

[0132] Step 5: Input the obtained segmented image regions into the image processing link respectively, obtain the regional enhanced images of the corresponding segmented image regions according to the image processing link model, merge the regional enhanced images, and obtain the adaptive enhanced image corresponding to the image data.

[0133] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An image adaptive enhancement system based on deep learning, including an image processing platform, characterized in that: The image processing platform is provided with an image acquisition module, an image processing module, an image analysis module, a model management module and an image management module; The image acquisition module is used to input image data and corresponding image information, generate image data packets and send them to the image processing module; The image processing module is used to perform region classification on the corresponding image according to the image data and the image information, perform image segmentation according to the region classification result, and obtain image quality data of the segmented image region and the corresponding pixel points; The image analysis module is used to perform classification processing according to the image quality data corresponding to the pixel points in the segmented image area, and obtain the comprehensive classification results of the pixel points corresponding to the segmented image area; The model management module is used to set image processing sub-models according to different comprehensive classification results, comprehensively sort the image processing sub-models, and construct image processing links and corresponding image processing link models; The image management module is used to input the obtained segmented image regions into the image processing link respectively, obtain the regional enhanced images of the corresponding segmented image regions according to the image processing link model, merge the regional enhanced images, and obtain the adaptive enhanced image corresponding to the image data.

2. The image adaptive enhancement system based on deep learning according to claim 1, characterized in that: The process of the image acquisition module recording image data and corresponding image information includes: Set up image input terminal and information input terminal; The image input terminal is used to input image data and mark the image data; the information input terminal is used to input image information corresponding to the image data, and the image information includes image description information and image key information; The obtained image data and image information are used to generate an image data packet according to the marking processing result.

3. The image adaptive enhancement system based on deep learning according to claim 2, characterized in that: The process of the image processing module obtaining the image quality data of the segmented image area and the corresponding pixel points includes: Obtaining pixel values ​​corresponding to corresponding pixel points according to the image data, obtaining a grayscale image corresponding to the image data according to the pixel values, performing edge detection processing on the grayscale image to determine corresponding edge pixel points, and generating an edge image according to position marks of the edge pixel points in the grayscale image; Setting a semantic analysis algorithm, analyzing and processing the image information through the semantic analysis algorithm, extracting corresponding descriptive semantic information and feature semantic information, comparing and analyzing the descriptive semantic information and feature semantic information with corresponding pixel points in the image data, determining corresponding pixel point positions, and marking the descriptive semantic information and feature semantic information according to the pixel point positions; According to the distribution of the corresponding edge pixel points in the edge image, the image sub-region is set, and according to the distribution results of the description semantic information and the feature semantic information in each image sub-region, it is determined whether to segment and extract the corresponding image sub-region, and the segmented image region after segmentation and extraction is obtained; Mark the segmented image area as the original image, set a mapping function, scale the pixel values ​​of the corresponding pixel points of the original image based on the mapping function, and obtain the corresponding scaled image; According to the mean square error of pixel values ​​between the original image and the scaled image, it is marked as image quality data.

4. The deep learning-based image adaptive enhancement system according to claim 3, characterized in that: The process of the image analysis module obtaining the comprehensive classification results of the pixels corresponding to the segmented image area includes: Obtain the distribution of pixel points corresponding to the description semantic information and the feature semantic information in the segmented image area, and divide the corresponding pixel points into key pixel points, basic pixel points and auxiliary pixel points according to the marking results of the corresponding pixel points; Acquire image quality data and historical image data packets within the segmented image area, preset pixel quality level intervals according to the historical image data packets, namely, high-quality pixel intervals, medium-quality pixel intervals, and low-quality pixel intervals, compare and analyze the image quality data with the corresponding pixel quality level intervals, and determine the pixel quality level to which the pixel point corresponding to the corresponding image quality data belongs according to the comparison and analysis results; The classification results corresponding to the above two different classification methods are obtained, and the obtained classification results are marked as the comprehensive classification results of the corresponding pixel points.

5. The deep learning-based image adaptive enhancement system according to claim 4, characterized in that: The process of the model management module setting the image processing sub-model according to different comprehensive classification results includes: Obtain the image processing process of the image data pixels corresponding to different comprehensive classification results, and generate image processing data sets corresponding to the corresponding types of comprehensive classification results; Based on the deep learning algorithm, the obtained image processing data set is analyzed and trained to construct an image processing sub-model corresponding to the corresponding type of comprehensive classification results.

6. The deep learning-based image adaptive enhancement system according to claim 5, characterized in that: The process of the model management module constructing the image processing link and the corresponding image processing link model includes: Obtaining the importance level of the pixel points corresponding to each comprehensive classification result in the image processing process, sorting the corresponding comprehensive classification results according to the importance level, setting the corresponding image processing space in sequence according to the sorting results, and correlating the image processing space with the corresponding image processing sub-model, inputting the pixel point area corresponding to the corresponding comprehensive classification result into the image processing space, connecting the various image processing spaces in series, and generating an image processing chain; The image merging process of each image processing space in the image processing link is obtained to generate a link data set. The constructed link data set is analyzed and processed based on the deep learning algorithm to construct an image processing link model.

7. The deep learning-based image adaptive enhancement system according to claim 6, characterized in that: The process of the image management module acquiring the adaptive enhanced image corresponding to the image data includes: Obtain the comprehensive classification result of the corresponding pixel point area in the segmented image area, input it into the image processing link, and the corresponding image processing sub-models in each image processing space in the image processing link analyze and process the corresponding pixel point area in turn until the enhanced analysis of the corresponding segmented image area is completed in the image processing link, and input it into the image processing link model for analysis and processing, and the corresponding segmented image area is enhanced image; The edge pixel points corresponding to the regional enhanced images are obtained, and the regional enhanced images are merged according to the distribution of the edge pixel points of each regional enhanced image to obtain an adaptive enhanced image corresponding to the corresponding image data.

8. The deep learning-based image adaptive enhancement method corresponding to the deep learning-based image adaptive enhancement system according to any one of claims 1 to 7, characterized in that: The following steps are involved: Step 1: Setting up an image processing platform, inputting image data and corresponding image information, and generating an image data packet according to the input image data and image information; Step 2: Perform regional classification on the corresponding image according to the image data and image information respectively, perform image segmentation according to the regional classification result, and obtain image quality data of the segmented image area and the corresponding pixel points; Step 3: Perform classification processing based on the image quality data of the corresponding pixels in the segmented image area to obtain the comprehensive classification results of the corresponding pixels in the segmented image area; Step 4: Set up image processing sub-models according to different comprehensive classification results, comprehensively sort the image processing sub-models, and construct image processing links and corresponding image processing link models; Step 5: Input the obtained segmented image regions into the image processing link respectively, obtain the regional enhanced images of the corresponding segmented image regions according to the image processing link model, merge the regional enhanced images, and obtain the adaptive enhanced image corresponding to the image data.

Images