Search

CN-121073833-B - Intelligent image recognition and restoration processing system and method based on big data

CN121073833BCN 121073833 BCN121073833 BCN 121073833BCN-121073833-B

Abstract

The invention relates to the field of image restoration, in particular to an intelligent image recognition restoration processing system and method based on big data, comprising the following steps: the system comprises an image verification module, a selection restoration module, a gray restoration module, a color separation enhancement module and a color mixing filter module, wherein the image verification module is used for determining the loss of image information, the selection restoration module is used for calculating the peak signal to noise ratio of an image, the gray restoration module is used for extending the gray texture of the image outwards, the color separation enhancement module is used for carrying out image color separation by adopting different color separation schemes, and the color mixing filter module is used for carrying out image denoising and image color mixing processing through a filter.

Inventors

  • JIANG GUZHENG

Assignees

  • 南京市乐知信息技术有限公司

Dates

Publication Date
20260508
Application Date
20250820

Claims (10)

  1. 1. The intelligent image recognition and restoration processing method based on big data is characterized by comprising the following steps of: S1, when an image is transmitted by a network, a source host calculates the number of image frames and the information quantity of each frame, a check code containing the integrity information of the image is generated, and the source host determines the information quantity loss in the image transmission process according to the check code; S2, mapping the data loss into equivalent noise power, calculating peak signal-to-noise ratio of the image, determining maximum allowable delay of image transmission according to network state, dynamically adjusting peak signal-to-noise ratio threshold, and selecting an image with peak signal-to-noise ratio lower than the threshold for restoration; s3, graying the image, calculating an isotopy line for the gray image, extracting a gray boundary, adopting an anisotropic diffusion model, extending the image isotopy line or the gray texture aligned with the adjacent frame outwards until encountering the gray boundary, and applying a diffusion result in the original image to obtain an enhanced image; S4, carrying out color separation on each pixel point in the enhanced image by adopting different color separation schemes, eliminating the correlation among color components through KL transformation, and determining the information quantity of each color component according to the information quantity index of each KL component; S5, selecting a filter with corresponding depth according to the information quantity of each color to perform denoising, weighting and fusing all color components according to the whole image information entropy proportion of each color separation scheme, performing pixel color mixing treatment, enabling the RGB channel information entropy of the repaired image to be consistent with the information entropy of the original image, and outputting the repaired image.
  2. 2. The intelligent image recognition and restoration processing method based on big data as set forth in claim 1, wherein the step S1 includes: s11, when an image stream or video is transmitted through a network, controlling an information source host to capture the image, dividing an image set into frames with fixed sizes, and calculating the information quantity of each frame of image based on entropy or discrete cosine transform DCT coefficients; s12, for each frame of picture, generating a check code containing image integrity information, wherein the check code comprises a frame ID, a frame size, information quantity, a hash value and redundant transmission codes, packaging the check code as a part of metadata in a picture data packet, sending the picture data packet to a host through a TCP or QUIC protocol, determining lost data of the picture by the host through comparing the block check codes, and calculating information quantity loss of the lost data.
  3. 3. The intelligent image recognition and restoration processing method based on big data as set forth in claim 2, wherein the step S2 includes: S21, complementing the missing region of the picture by using a pre-trained deep learning model, calculating local PSNR values of all color channels of the picture, and weighting according to the sensitivity of human eyes to obtain the peak signal-to-noise ratio of the picture; S22, detecting network transmission states including transmission speed, delay and jitter, setting maximum allowable image transmission delay according to the network states, calculating a dynamic restoration threshold value, enabling restoration information quantity/average restoration speed of all images under the restoration threshold value to be smaller than the maximum allowable delay, and selecting images with peak signal-to-noise ratio lower than the threshold value as image frames needing restoration.
  4. 4. The intelligent image recognition and restoration processing method based on big data as set forth in claim 3, wherein the step S3 includes: S31, generating a gray image by adopting a weighted graying formula, extracting a gray boundary by a Canny operator or a deep learning edge detection model, detecting a weak edge by a Sobel operator to form a complete gray boundary mask, and generating an isotopy line with single pixel width according to the mask; S32, aligning the isotopy lines of the adjacent frames, adopting an anisotropic diffusion model, and based on a Perona-Malik diffusion equation, diffusing and extending the image isotopy lines or gray textures aligned with the isotopy lines of the adjacent frames outwards, and stopping diffusion when encountering a boundary mask to generate a gray-scale diffused image; and S33, reserving chromaticity information of the original image, applying a diffusion result to the brightness channel, and restoring the gray level image into a colored image to obtain an enhanced image.
  5. 5. The intelligent image recognition and restoration processing method based on big data as set forth in claim 4, wherein the step S4 includes: S41, determining different space color separation schemes, wherein the color separation schemes comprise RGB, HSV and Lab color separation, independently calculating covariance matrixes for color components of each color separation scheme, decomposing eigenvalues to determine a transformation matrix, performing KL transformation based on the transformation matrix, and performing decorrelation on each color component of a picture; S42, reserving the first k eigenvectors of the transformation matrix by taking eigenvalues and variances of all KL components as information quantity indexes, determining component duty ratios of all pixel points according to variances of color components corresponding to the eigenvectors, calculating local entropy of each color component, and accumulating according to a component duty ratio full graph to obtain information entropy of each color component; the step S5 comprises the following steps: S51, selecting a color separation scheme to maximize the full-image information entropy for each pixel point, and selecting a small-radius bilateral filter or a Gaussian filter with corresponding depth according to the information quantity of each color to denoise the image; S52, weighting and fusing all color components according to the whole image information entropy proportion of each color separation scheme, carrying out pixel color mixing treatment under different color separation schemes, enabling the RGB channel information entropy of the restored image to be consistent with the information entropy of the original image, and outputting the restored image.
  6. 6. The intelligent image recognition and restoration processing system based on big data is characterized by comprising an image verification module, a selection restoration module, a gray level restoration module, a color separation enhancement module and a color mixing filtering module; The image verification module is used for controlling the information source host to divide an image into frames with fixed size when an image stream or a video is transmitted through a network, verifying the number of frames of the image and the information quantity of the image, and generating a verification code containing the integrity information of the image, wherein the verification code comprises a frame ID, the frame size, the information quantity, a hash value and a redundant transmission code, and the information source host detects data loss according to the verification code and determines the information quantity loss in the image transmission process; The selection restoration module is used for complementing the image missing region by using a pre-trained deep learning model, mapping the data loss amount into equivalent noise power, respectively calculating local signal-to-noise ratios of all channels of the image, weighting according to human eye sensitivity to obtain peak signal-to-noise ratios of the image, determining the maximum allowable delay of image transmission according to the network state, dynamically adjusting the peak signal-to-noise ratio threshold, and selecting an image with the peak signal-to-noise ratio lower than the threshold for restoration; The gray repair module is used for graying an image, extracting a gray boundary by using a Canny operator or a deep learning edge detection model, calculating an isoperimetric line for the gray image, adopting an anisotropic diffusion model, using a gradient diffusion algorithm to diffuse and extend the gray texture of the image isoperimetric line or an aligned adjacent frame outwards until encountering the gray boundary, generating a gray diffused image, retaining the chromaticity information of an original image, and applying a diffusion result in the original image to obtain an enhanced image; The color separation enhancement module is used for carrying out color separation on each pixel point in the enhanced image by adopting different color separation schemes based on local color clustering, independently calculating a covariance matrix for color components of each color separation scheme, eliminating correlation among the color components through KL transformation, and calculating characteristic values and variances of the KL components as information quantity indexes to determine the information quantity of colors of each channel; The color mixing filtering module is used for selecting a color separation scheme to maximize the full-image information entropy for each pixel point, selecting a small-radius bilateral filter or a Gaussian filter with corresponding depth according to the information quantity of each color to perform image denoising, weighting and fusing the filtered color separation scheme results according to the information quantity, carrying out color mixing treatment on the pixels under different color separation schemes, keeping the RGB channel information entropy of the restored image consistent with the information entropy of the original image, and outputting the restored image.
  7. 7. The intelligent image recognition and restoration processing system based on big data as set forth in claim 6, wherein the image verification module comprises an information entropy accounting unit and an integrity verification unit; The information entropy calculation unit is used for capturing images and calculating the information quantity of each frame of image based on entropy or discrete cosine transform DCT coefficients; The integrity check unit is used for generating a check code as metadata, packaging the check code into a picture data packet, and determining picture lost data by comparing the block check codes; The selection repair module comprises a peak noise unit and a dynamic selection unit; The peak noise unit is used for calculating local PSNR values of all color channels of the picture, and obtaining the peak signal-to-noise ratio of the picture by weighted average; The dynamic selection unit is used for detecting network transmission states including transmission speed, delay and jitter, calculating a dynamic restoration threshold value and determining picture frames needing restoration.
  8. 8. The intelligent image recognition and restoration processing system based on big data as set forth in claim 7, wherein the gray scale restoration module comprises a boundary recognition unit, a gradient diffusion unit and a gray scale restoration unit; the boundary recognition unit is used for generating a gray image by adopting a weighted graying formula, detecting a weak edge by using a Sobel operator, and detecting a complete gray boundary mask; The gradient diffusion unit is used for generating an isocenter line with single pixel width, and performing texture diffusion along the direction of the isocenter line based on a Perona-Malik diffusion equation; the gray-scale restoring unit is used for aligning the isotopy lines of adjacent frames and restoring the gray-scale image into a colored image.
  9. 9. The intelligent image recognition and restoration processing system based on big data as set forth in claim 8, wherein the color separation enhancement module comprises a color separation unit and a correlation elimination unit; The color separation unit is used for determining different space color separation schemes, including RGB, HSV and Lab color separation; the correlation removal unit de-correlates each color component of the picture by KL transform and calculates information amounts, respectively.
  10. 10. The intelligent image recognition and restoration processing system based on big data as set forth in claim 9, wherein the color mixing filtering module comprises a selection filtering unit, a color mixing unit and an information restoring unit; The selection filtering unit is used for dynamically selecting a filter with a corresponding depth according to the color separation information quantity to perform monochromatic denoising operation; The color mixing unit is used for weighting and fusing color components of each color separation scheme according to the information entropy proportion, and restoring an original image; the information recovery unit is used for verifying the information entropy of the original image and the repair image and performing fine adjustment on the image area with the difference.

Description

Intelligent image recognition and restoration processing system and method based on big data Technical Field The invention relates to the field of image restoration, in particular to an intelligent image recognition restoration processing system and method based on big data. Background Image restoration is a technique for restoring or filling a missing or damaged portion of an image using a computer algorithm, with the aim of visually restoring and reconstructing the missing region of the image or improving the resolution of the image lost during transmission. Common image restoration methods include diffusion synthesis and texture synthesis, and similar texture restoration image loss is diffused through pixel information, and in recent years, deep learning restoration techniques based on CNN or GAN are also in development and application. Although the neural network is adopted to repair the image, the method can obtain a repair result with higher precision, but has the advantages of large model calculation amount, high data demand and low repair speed, and can inevitably cause transmission delay in high-speed image transmission scenes such as image streaming or video, and is difficult to apply in a large scale in real time. When the traditional texture diffusion technology pursues efficient restoration, the problems of blurring of the restored image texture, structural distortion and the like are easy to generate, the controllability of restoration of the image color is insufficient, and the phenomenon of pseudo color is easy to occur. In addition, in real-time network interaction, the main goal of image restoration is to reduce the loss of images in the data exchange process so as to obtain a better ornamental effect, so that the selective restoration of images and information matching have great influence on user experience, and the actual restoration needs can be better met by integral optimization. Disclosure of Invention The invention aims to provide an intelligent image recognition and restoration processing system and method based on big data, which are used for solving the problems in the background technology. In order to solve the technical problems, the invention provides a technical scheme that an intelligent image recognition and restoration processing system based on big data comprises an image verification module, a selection restoration module, a gray level restoration module, a color separation enhancement module and a color mixing filtering module; The image verification module is used for controlling the information source host to divide an image into frames with fixed size when an image stream or a video is transmitted through a network, verifying the number of frames of the image and the information quantity of the image, and generating a verification code containing the integrity information of the image, wherein the verification code comprises a frame ID, the frame size, the information quantity, a hash value and a redundant transmission code, and the information source host detects data loss according to the verification code and determines the information quantity loss in the image transmission process; The selection restoration module is used for complementing the image missing region by using a pre-trained deep learning model, mapping the data loss amount into equivalent noise power, respectively calculating local signal-to-noise ratios of all channels of the image, weighting according to human eye sensitivity to obtain peak signal-to-noise ratios of the image, determining the maximum allowable delay of image transmission according to the network state, dynamically adjusting the peak signal-to-noise ratio threshold, and selecting an image with the peak signal-to-noise ratio lower than the threshold for restoration; The gray repair module is used for graying an image, extracting a gray boundary by using a Canny operator or a deep learning edge detection model, calculating an isoperimetric line for the gray image, adopting an anisotropic diffusion model, using a gradient diffusion algorithm to diffuse and extend the gray texture of the image isoperimetric line or an aligned adjacent frame outwards until encountering the gray boundary, generating a gray diffused image, retaining the chromaticity information of an original image, and applying a diffusion result in the original image to obtain an enhanced image; The color separation enhancement module is used for carrying out color separation on each pixel point in the enhanced image by adopting different color separation schemes based on local color clustering, independently calculating a covariance matrix for color components of each color separation scheme, eliminating correlation among the color components through KL transformation, and calculating characteristic values and variances of the KL components as information quantity indexes to determine the information quantity of colors of each channel; The color mixing filtering modu