An intelligent image recognition repair processing system and method based on big data

Abstract

The application relates to the field of image restoration, in particular to an intelligent image recognition and restoration processing system and method based on big data, which comprises an image verification module, a selection restoration module, a gray scale restoration module, a color separation enhancement module and a color mixing filter module; the image verification module is used for determining image information loss; the selection restoration module is used for calculating the peak signal-to-noise ratio of the image; the gray scale restoration module is used for outwardly extending the gray scale texture of the image; the color separation enhancement module is used for separating the image colors by using 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. The application can avoid the data block effect, blurring and color distortion caused by network fluctuation, reduce the image restoration time, adapt to the viewing demand of users, reduce the influence of data loss on the image, optimize the calculation efficiency of the restoration algorithm, improve the picture texture visual sense and improve the visual experience of terminal users.

Description

Technical Field

[0001] This invention relates to the field of image restoration, specifically to an intelligent image recognition and restoration processing system and method based on big data. Background Technology

[0002] Image inpainting is a technique that uses computer algorithms to repair or fill in missing or damaged parts of an image. Its purpose is to visually restore and reconstruct missing areas of an image or improve the resolution lost during transmission. Common image inpainting methods include diffusion synthesis and texture synthesis, which use pixel information to diffuse similar textures to repair image loss. In recent years, deep learning inpainting techniques based on CNNs or GANs have also been developed and applied.

[0003] While neural networks can achieve more accurate image inpainting results, they are computationally intensive, require a lot of data, and are slow. In high-speed image transmission scenarios such as image streams or videos, this inevitably causes transmission delays, making real-time large-scale applications difficult. Traditional texture diffusion techniques, while aiming for efficient inpainting, easily lead to problems such as blurred textures or structural distortion in the repaired image. They also lack control over image color restoration, easily resulting in false-color phenomena.

[0004] Furthermore, in real-time network interaction, the main goal of image restoration is to reduce image loss during data exchange in order to achieve a better viewing experience. Therefore, selective image restoration and information matching have a significant impact on user experience and require overall optimization to better meet actual restoration needs. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent image recognition and repair processing system and method based on big data, so as to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent image recognition and repair processing system based on big data, comprising: an image verification module, a selection repair module, a grayscale repair module, a color separation enhancement module, and a color mixing filtering module;

[0007] The image verification module is used to control the source host to divide the image into fixed-size frames when transmitting image streams or videos over the network, calculate the number of image frames and the amount of image information, and generate a verification code containing image integrity information. The verification code includes: frame ID, frame size, amount of information, hash value and redundant transmission code. The source host detects data loss based on the verification code and determines the amount of information lost during image transmission.

[0008] The selection and repair module is used to fill in the missing areas of the image using a pre-trained deep learning model, map the data loss to the equivalent noise power, calculate the local signal-to-noise ratio of each channel of the image, weight it according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image, determine the maximum allowable delay of image transmission according to the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair.

[0009] The grayscale restoration module is used to convert the image to grayscale, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, calculate isoluminance lines for the grayscale image, and use an anisotropic diffusion model and gradient diffusion algorithm to diffuse and extend the isoluminance lines or grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries, generating a grayscale diffused image while retaining the chromaticity information of the original image. The diffusion results are then applied to the original image to obtain an enhanced image.

[0010] The color separation enhancement module is used to perform color separation for each pixel in the enhanced image based on local color clustering and different color separation schemes. The covariance matrix of each color component of the color separation scheme is calculated separately. The correlation between each color component is eliminated by KL transformation. The eigenvalues ​​and variances of each KL component are calculated as information content indicators to determine the information content of each channel color.

[0011] The color mixing and filtering module is used to select a color separation scheme for each pixel to maximize the information entropy of the whole image. It selects a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising. It then weights and fuses the results of each color separation scheme after filtering according to the information content, performs pixel color mixing processing under different color separation schemes, and makes the information entropy of the RGB channels of the restored image consistent with the information entropy of the original image, and outputs the restored image.

[0012] Furthermore, the image verification module includes: an information entropy calculation unit and an integrity verification unit;

[0013] The information entropy calculation unit is used to capture images and calculate the information content of each frame of image based on entropy or discrete cosine transform (DCT) coefficients.

[0014] The integrity verification unit is used to generate a check code as metadata encapsulated in the image data packet, and to determine the missing data in the image by comparing the block check codes.

[0015] Furthermore, the selection and repair module includes: a peak noise unit and a dynamic selection unit;

[0016] The peak noise unit is used to calculate the local PSNR value of each color channel of the image, and the weighted average is used to obtain the peak signal-to-noise ratio of the image.

[0017] The dynamic selection unit is used to detect network transmission status, including transmission speed, latency and jitter, calculate dynamic repair threshold, and determine the image frames that need to be repaired.

[0018] Furthermore, the grayscale restoration module includes: a boundary recognition unit, a gradient diffusion unit, and a grayscale restoration unit;

[0019] The boundary recognition unit is used to generate a grayscale image using a weighted grayscale formula, detect weak edges using the Sobel operator, and detect a complete grayscale boundary mask.

[0020] The gradient diffusion unit is used to generate iso-illuminance lines with a width of one pixel, and performs texture diffusion along the iso-illuminance line direction based on the Perona-Malik diffusion equation.

[0021] The grayscale restoration unit is used to align the isoluminance lines of adjacent frames and restore the grayscale image to a colored image.

[0022] Furthermore, the color separation enhancement module includes: a color separation unit and a correlation elimination unit;

[0023] The color separation unit is used to determine different color separation schemes, including RGB, HSV and Lab color separation;

[0024] The correlation elimination unit performs KL transform to decorrelate each color component of the image and calculates the information content for each component.

[0025] Furthermore, the color mixing and filtering module includes: a selection filtering unit, a color mixing unit, and an information recovery unit;

[0026] The selection filtering unit is used to dynamically select the filter of the corresponding depth according to the amount of color separation information to perform monochrome noise reduction operation;

[0027] The color mixing unit is used to weight and fuse the color components of each color separation scheme according to the information entropy ratio to restore the original image;

[0028] The information restoration unit is used to verify the information entropy of the original image and the restored image, and to fine-tune the image regions where there are differences.

[0029] A big data-based intelligent image recognition and restoration method includes the following steps:

[0030] Step S1. When transmitting images over the network, the source host calculates the number of image frames and the amount of information in each frame, generates a check code containing image integrity information, and the source host determines the amount of information lost during image transmission based on the check code.

[0031] Step S2. Map the data loss to equivalent noise power, calculate the peak signal-to-noise ratio of the image, determine the maximum allowable delay for image transmission based on the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair.

[0032] Step S3. Convert the image to grayscale, calculate the isoluminance lines for the grayscale image, extract the grayscale boundaries, and use an anisotropic diffusion model to extend the isoluminance lines of the image or the grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries. Apply the diffusion results to the original image to obtain the enhanced image.

[0033] Step S4. For each pixel in the enhanced image, different color separation schemes are used for color separation. The correlation between each color component is eliminated by KL transformation. The information content of each color component is determined according to the information content index of each KL component.

[0034] Step S5. Select the corresponding depth filter for noise reduction according to the information content of each color, and weight and fuse all color components according to the full image information entropy ratio of each color separation scheme. Perform pixel color mixing to make the RGB channel information entropy of the repaired image consistent with the information entropy of the original image, and output the repaired image.

[0035] Furthermore, step S1 includes:

[0036] Step S11. When transmitting image streams or videos over a network, control the source host to capture images, divide the image set into frames of fixed size, and calculate the information content of each frame based on entropy or Discrete Cosine Transform (DCT) coefficients.

[0037] Step S12. For each frame of image, generate a check code containing image integrity information. The check code includes: frame ID, frame size, information content, hash value, and redundant transmission code. The check code is encapsulated as part of the metadata in the image data packet and sent to the host machine via TCP or QUIC protocol. The host machine determines the lost image data by comparing the block check codes and calculates the information content loss of the lost data.

[0038] Furthermore, step S2 includes:

[0039] Step S21. Use a pre-trained deep learning model to fill in the missing areas of the image, calculate the local PSNR values ​​of each color channel of the image, and weight them according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image;

[0040] Step S22. Detect the network transmission status, including transmission speed, latency and jitter. Set the maximum allowable delay for image transmission based on the network status. Calculate the dynamic repair threshold so that the repair information content / average repair speed of all images under the repair threshold is less than the maximum allowable delay. Select images with a peak signal-to-noise ratio lower than the threshold as the image frames that need to be repaired.

[0041] Furthermore, step S3 includes:

[0042] Step S31. Generate a grayscale image using a weighted grayscale formula, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, detect weak edges using the Sobel operator, form a complete grayscale boundary mask, and generate iso-illuminance lines with a single pixel width based on the mask.

[0043] Step S32. Align the isoluminance lines of adjacent frames, adopt the anisotropic diffusion model, based on the Perona-Malik diffusion equation, and diffuse and extend the grayscale texture of the image isoluminance lines or the isoluminance lines of adjacent frames outward. The diffusion terminates when it encounters the boundary mask, generating the grayscale diffused image.

[0044] Step S33. Preserve the chromaticity information of the original image, apply the diffusion result to the luminance channel, and restore the grayscale image to a colored image to obtain the enhanced image.

[0045] Furthermore, step S4 includes:

[0046] Step S41. Determine different color separation schemes, including RGB, HSV and Lab color separation. Calculate the covariance matrix for each color component of the color separation scheme separately. Determine the transformation matrix by eigenvalue decomposition. Perform KL transformation based on the transformation matrix to decorrelate each color component of the image.

[0047] Step S42. The eigenvalues ​​and variances of each KL component are used as information content indicators. The first k eigenvectors of the transformation matrix are retained. The component proportion of each pixel is determined by the variance of the color component corresponding to the eigenvector. k is the effective information dimension. The local entropy of each color component is calculated. After accumulating the entropy of each color component across the entire image according to the component proportion, the information entropy of each color component is obtained.

[0048] Furthermore, step S5 includes:

[0049] Step S51. For each pixel, select a color separation scheme to maximize the information entropy of the whole image, and select a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising.

[0050] Step S52. Weight and fuse all color components according to the full image information entropy ratio of each color separation scheme, perform pixel color mixing processing under different color separation schemes, so that the RGB channel information entropy of the restored image is consistent with the information entropy of the original image, and output the restored image.

[0051] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0052] 1. This invention can calculate the number of image frames and the amount of image information when transmitting images over a network, treat the amount of data loss as noise, determine the peak signal-to-noise ratio of the image, and perform image restoration when the peak signal-to-noise ratio is lower than a threshold. This allows the sending end to use a higher compression ratio to transmit images, reduces transmission pressure, avoids data block effects, blurring and color distortion caused by network fluctuations, reduces image restoration time, and adapts to user viewing needs.

[0053] 2. This invention converts an image to grayscale and uses a gradient diffusion algorithm to extend the grayscale texture outward to the grayscale boundary. Based on the original image, color restoration is performed, which can repair color mapping, improve the texture performance displayed on different devices, reduce the impact of data loss on the image, restore image details, improve the image texture appearance, and significantly enhance the visual experience of end users.

[0054] 3. This invention employs different color separation schemes for color separation, eliminates the correlation between color components through KL transform to determine the amount of color information, selects filters of corresponding depths according to the amount of color information for image restoration, and performs color mixing processing to ensure that the amount of information in the restored image is the same as that in the original image. It can optimize the processing effect for different scenarios and restoration targets, specifically repair different image defects, optimize the computational efficiency of the restoration algorithm, adapt to different degradation types, and improve the quality of color restoration. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a schematic diagram of the structure of an intelligent image recognition and repair processing system based on big data according to the present invention;

[0057] Figure 2 This is a schematic diagram illustrating the steps of an intelligent image recognition and repair processing method based on big data according to the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figure 1 The present invention provides a technical solution: an intelligent image recognition and repair processing system based on big data, comprising: an image verification module, a selection repair module, a grayscale repair module, a color separation enhancement module, and a color mixing filtering module;

[0060] The image verification module is used to control the source host to divide the image into fixed-size frames when transmitting image streams or videos over the network, calculate the number of image frames and the amount of image information, and generate a verification code containing image integrity information. The verification code includes: frame ID, frame size, amount of information, hash value and redundant transmission code. The source host detects data loss based on the verification code and determines the amount of information lost during image transmission.

[0061] The image verification module includes: an information entropy calculation unit and an integrity verification unit;

[0062] The information entropy calculation unit is used to capture images and calculate the information content of each frame of image based on entropy or discrete cosine transform (DCT) coefficients.

[0063] The integrity verification unit is used to generate a check code as metadata encapsulated in the image data packet, and to determine the missing data in the image by comparing the block check codes.

[0064] The selection and repair module is used to fill in the missing areas of the image using a pre-trained deep learning model, map the data loss to the equivalent noise power, calculate the local signal-to-noise ratio of each channel of the image, weight it according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image, determine the maximum allowable delay of image transmission according to the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair.

[0065] The selection and repair module includes: a peak noise unit and a dynamic selection unit;

[0066] The peak noise unit is used to calculate the local PSNR value of each color channel of the image, and the weighted average is used to obtain the peak signal-to-noise ratio of the image.

[0067] The dynamic selection unit is used to detect network transmission status, including transmission speed, latency and jitter, calculate dynamic repair threshold, and determine the image frames that need to be repaired.

[0068] The grayscale restoration module is used to convert the image to grayscale, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, calculate isoluminance lines for the grayscale image, and use an anisotropic diffusion model and gradient diffusion algorithm to diffuse and extend the isoluminance lines or grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries, generating a grayscale diffused image while retaining the chromaticity information of the original image. The diffusion results are then applied to the original image to obtain an enhanced image.

[0069] The grayscale restoration module includes: a boundary recognition unit, a gradient diffusion unit, and a grayscale restoration unit;

[0070] The boundary recognition unit is used to generate a grayscale image using a weighted grayscale formula, detect weak edges using the Sobel operator, and detect a complete grayscale boundary mask.

[0071] The gradient diffusion unit is used to generate iso-illuminance lines with a width of one pixel, and performs texture diffusion along the iso-illuminance line direction based on the Perona-Malik diffusion equation.

[0072] The grayscale restoration unit is used to align the isoluminance lines of adjacent frames and restore the grayscale image to a colored image.

[0073] The color separation enhancement module is used to perform color separation for each pixel in the enhanced image based on local color clustering and different color separation schemes. The covariance matrix of each color component of the color separation scheme is calculated separately. The correlation between each color component is eliminated by KL transformation. The eigenvalues ​​and variances of each KL component are calculated as information content indicators to determine the information content of each channel color.

[0074] The color separation enhancement module includes: a color separation unit and a correlation elimination unit;

[0075] The color separation unit is used to determine different color separation schemes, including RGB, HSV and Lab color separation;

[0076] The correlation elimination unit performs KL transform to decorrelate each color component of the image and calculates the information content for each component.

[0077] The color mixing and filtering module is used to select a color separation scheme for each pixel to maximize the information entropy of the whole image. It selects a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising. It then weights and fuses the results of each color separation scheme after filtering according to the information content, performs pixel color mixing processing under different color separation schemes, and makes the information entropy of the RGB channels of the restored image consistent with the information entropy of the original image, and outputs the restored image.

[0078] The color mixing and filtering module includes: a selection filtering unit, a color mixing unit, and an information recovery unit;

[0079] The selection filtering unit is used to dynamically select the filter of the corresponding depth according to the amount of color separation information to perform monochrome noise reduction operation;

[0080] The color mixing unit is used to weight and fuse the color components of each color separation scheme according to the information entropy ratio to restore the original image;

[0081] The information restoration unit is used to verify the information entropy of the original image and the restored image, and to fine-tune the image regions where there are differences.

[0082] like Figure 2 As shown, an intelligent image recognition and restoration method based on big data includes the following steps:

[0083] Step S1. When transmitting images over the network, the source host calculates the number of image frames and the amount of information in each frame, generates a check code containing image integrity information, and the source host determines the amount of information lost during image transmission based on the check code.

[0084] Step S1 includes:

[0085] Step S11. When transmitting image streams or videos over a network, control the source host to capture images, divide the image set into frames of fixed size, and calculate the information content of each frame based on entropy or Discrete Cosine Transform (DCT) coefficients.

[0086] Step S12. For each frame of image, generate a check code containing image integrity information. The check code includes: frame ID, frame size, information content, hash value, and redundant transmission code. The check code is encapsulated as part of the metadata in the image data packet and sent to the host machine via TCP or QUIC protocol. The host machine determines the lost image data by comparing the block check codes and calculates the information content loss of the lost data.

[0087] Step S2. Map the data loss to equivalent noise power, calculate the peak signal-to-noise ratio of the image, determine the maximum allowable delay for image transmission based on the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair.

[0088] Step S2 includes:

[0089] Step S21. Use a pre-trained deep learning model to fill in the missing areas of the image, calculate the local PSNR values ​​of each color channel of the image, and weight them according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image;

[0090] Step S22. Detect the network transmission status, including transmission speed, latency and jitter. Set the maximum allowable delay for image transmission based on the network status. Calculate the dynamic repair threshold so that the repair information content / average repair speed of all images under the repair threshold is less than the maximum allowable delay. Select images with a peak signal-to-noise ratio lower than the threshold as the image frames that need to be repaired.

[0091] Step S3. Convert the image to grayscale, calculate the isoluminance lines for the grayscale image, extract the grayscale boundaries, and use an anisotropic diffusion model to extend the isoluminance lines of the image or the grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries. Apply the diffusion results to the original image to obtain the enhanced image.

[0092] Step S3 includes:

[0093] Step S31. Generate a grayscale image using a weighted grayscale formula, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, detect weak edges using the Sobel operator, form a complete grayscale boundary mask, and generate iso-illuminance lines with a single pixel width based on the mask.

[0094] Step S32. Align the isoluminance lines of adjacent frames, adopt the anisotropic diffusion model, based on the Perona-Malik diffusion equation, and diffuse and extend the grayscale texture of the image isoluminance lines or the isoluminance lines of adjacent frames outward. The diffusion terminates when it encounters the boundary mask, generating the grayscale diffused image.

[0095] Step S33. Preserve the chromaticity information of the original image, apply the diffusion result to the luminance channel, and restore the grayscale image to a colored image to obtain the enhanced image.

[0096] Step S4. For each pixel in the enhanced image, different color separation schemes are used for color separation. The correlation between each color component is eliminated by KL transformation. The information content of each color component is determined according to the information content index of each KL component.

[0097] Step S4 includes:

[0098] Step S41. Determine different color separation schemes, including RGB, HSV and Lab color separation. Calculate the covariance matrix for each color component of the color separation scheme separately. Determine the transformation matrix by eigenvalue decomposition. Perform KL transformation based on the transformation matrix to decorrelate each color component of the image.

[0099] Step S42. The eigenvalues ​​and variances of each KL component are used as information content indicators. The first k eigenvectors of the transformation matrix are retained. The component proportion of each pixel is determined by the variance of the color component corresponding to the eigenvector. k is the effective information dimension. The local entropy of each color component is calculated. After accumulating the entropy of each color component across the entire image according to the component proportion, the information entropy of each color component is obtained.

[0100] Step S5. Select the corresponding depth filter for noise reduction according to the information content of each color, and weight and fuse all color components according to the full image information entropy ratio of each color separation scheme. Perform pixel color mixing to make the RGB channel information entropy of the repaired image consistent with the information entropy of the original image, and output the repaired image.

[0101] Step S5 includes:

[0102] Step S51. For each pixel, select a color separation scheme to maximize the information entropy of the whole image, and select a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising.

[0103] Step S52. Weight and fuse all color components according to the full image information entropy ratio of each color separation scheme, perform pixel color mixing processing under different color separation schemes, so that the RGB channel information entropy of the restored image is consistent with the information entropy of the original image, and output the restored image.

[0104] Example: A user plays a 1-second, 5-frame video on the host computer. After parsing the checksum, the information loss of the 5 images transmitted from the source host within 1 second is 20 bits, 10 bits, 15 bits, 18 bits, and 5 bits, respectively. The peak signal-to-noise ratios are 11.5, 20, 18, 15, and 40, respectively. The maximum allowable delay is 0.3 seconds, and the average repair speed is 10 bits / second. Therefore, the repair threshold is set to 15, and the 1st and 4th frames are repaired, parsing grayscale texture extension and color channel mixing.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0106] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent image recognition and restoration processing based on big data, characterized in that, The method includes the following steps: Step S1. When transmitting images over the network, the source host calculates the number of image frames and the amount of information in each frame, generates a check code containing image integrity information, and the source host determines the amount of information lost during image transmission based on the check code. Step S2. Map the data loss to equivalent noise power, calculate the peak signal-to-noise ratio of the image, determine the maximum allowable delay for image transmission based on the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair. Step S3. Convert the image to grayscale, calculate the isoluminance lines for the grayscale image, extract the grayscale boundaries, and use an anisotropic diffusion model to extend the isoluminance lines of the image or the grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries. Apply the diffusion results to the original image to obtain the enhanced image. Step S4. For each pixel in the enhanced image, different color separation schemes are used for color separation. The correlation between each color component is eliminated by KL transformation. The information content of each color component is determined according to the information content index of each KL component. Step S5. Select the corresponding depth filter for noise reduction according to the information content of each color, and weight and fuse all color components according to the full image information entropy ratio of each color separation scheme. Perform pixel color mixing to make the RGB channel information entropy of the repaired image consistent with the information entropy of the original image, and output the repaired image.

2. The intelligent image recognition and restoration processing method based on big data according to claim 1, characterized in that: Step S1 includes: Step S11. When transmitting image streams or videos over a network, control the source host to capture images, divide the image set into frames of fixed size, and calculate the information content of each frame based on entropy or Discrete Cosine Transform (DCT) coefficients. Step S12. For each frame of image, generate a check code containing image integrity information. The check code includes: frame ID, frame size, information content, hash value, and redundant transmission code. The check code is encapsulated as part of the metadata in the image data packet and sent to the host machine via TCP or QUIC protocol. The host machine determines the lost image data by comparing the block check codes and calculates the information content loss of the lost data.

3. The intelligent image recognition and restoration processing method based on big data according to claim 2, characterized in that: Step S2 includes: Step S21. Use a pre-trained deep learning model to fill in the missing areas of the image, calculate the local PSNR values ​​of each color channel of the image, and weight them according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image; Step S22. Detect the network transmission status, including transmission speed, latency and jitter. Set the maximum allowable delay for image transmission based on the network status. Calculate the dynamic repair threshold so that the repair information content / average repair speed of all images under the repair threshold is less than the maximum allowable delay. Select images with a peak signal-to-noise ratio lower than the threshold as the image frames that need to be repaired.

4. The intelligent image recognition and restoration processing method based on big data according to claim 3, characterized in that: Step S3 includes: Step S31. Generate a grayscale image using a weighted grayscale formula, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, detect weak edges using the Sobel operator, form a complete grayscale boundary mask, and generate iso-illuminance lines with a single pixel width based on the mask. Step S32. Align the isoluminance lines of adjacent frames, adopt the anisotropic diffusion model, based on the Perona-Malik diffusion equation, and diffuse and extend the grayscale texture of the image isoluminance lines or the isoluminance lines of adjacent frames outward. The diffusion terminates when it encounters the boundary mask, generating the grayscale diffused image. Step S33. Preserve the chromaticity information of the original image, apply the diffusion result to the luminance channel, and restore the grayscale image to a colored image to obtain the enhanced image.

5. The intelligent image recognition and restoration processing method based on big data according to claim 4, characterized in that: Step S4 includes: Step S41. Determine different color separation schemes, including RGB, HSV and Lab color separation. Calculate the covariance matrix for each color component of the color separation scheme separately. Determine the transformation matrix by eigenvalue decomposition. Perform KL transformation based on the transformation matrix to decorrelate each color component of the image. Step S42. The eigenvalues ​​and variances of each KL component are used as information content indicators. The first k eigenvectors of the transformation matrix are retained. The component proportion of each pixel is determined by the variance of the color component corresponding to the eigenvector. k is the effective information dimension. The local entropy of each color component is calculated. After the entropy of each color component is accumulated in the whole image according to the component proportion, the information entropy of each color component is obtained. Step S5 includes: Step S51. For each pixel, select a color separation scheme to maximize the information entropy of the whole image, and select a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising. Step S52. Weight and fuse all color components according to the full image information entropy ratio of each color separation scheme, perform pixel color mixing processing under different color separation schemes, so that the RGB channel information entropy of the restored image is consistent with the information entropy of the original image, and output the restored image.

6. A smart image recognition and restoration processing system based on big data, characterized in that, The system includes the following modules: image verification module, selection repair module, grayscale repair module, color separation enhancement module, and color mixing filtering module; The image verification module is used to control the source host to divide the image into fixed-size frames when transmitting image streams or videos over the network, calculate the number of image frames and the amount of image information, and generate a verification code containing image integrity information. The verification code includes: frame ID, frame size, amount of information, hash value and redundant transmission code. The source host detects data loss based on the verification code and determines the amount of information lost during image transmission. The selection and repair module is used to fill in the missing areas of the image using a pre-trained deep learning model, map the data loss to the equivalent noise power, calculate the local signal-to-noise ratio of each channel of the image, weight it according to human eye sensitivity to obtain the peak signal-to-noise ratio of the image, determine the maximum allowable delay of image transmission according to the network status, dynamically adjust the peak signal-to-noise ratio threshold, and select images with a peak signal-to-noise ratio lower than the threshold for repair. The grayscale restoration module is used to convert the image to grayscale, extract grayscale boundaries using the Canny operator or a deep learning edge detection model, calculate isoluminance lines for the grayscale image, and use an anisotropic diffusion model and gradient diffusion algorithm to diffuse and extend the isoluminance lines or grayscale textures aligned with adjacent frames outward until they encounter grayscale boundaries, generating a grayscale diffused image while retaining the chromaticity information of the original image. The diffusion results are then applied to the original image to obtain an enhanced image. The color separation enhancement module is used to perform color separation for each pixel in the enhanced image based on local color clustering and different color separation schemes. The covariance matrix of each color component of the color separation scheme is calculated separately. The correlation between each color component is eliminated by KL transformation. The eigenvalues ​​and variances of each KL component are calculated as information content indicators to determine the information content of each channel color. The color mixing and filtering module is used to select a color separation scheme for each pixel to maximize the information entropy of the whole image. It selects a small-radius bilateral filter or Gaussian filter of corresponding depth according to the information content of each color to perform image denoising. It then weights and fuses the results of each color separation scheme after filtering according to the information content, performs pixel color mixing processing under different color separation schemes, and makes the information entropy of the RGB channels of the restored image consistent with the information entropy of the original image, and outputs the restored image.

7. The intelligent image recognition and restoration processing system based on big data according to claim 6, characterized in that: The image verification module includes: an information entropy calculation unit and an integrity verification unit; The information entropy calculation unit is used to capture images and calculate the information content of each frame of image based on entropy or discrete cosine transform (DCT) coefficients. The integrity verification unit is used to generate a check code as metadata encapsulated in the image data packet, and to determine the missing data in the image by comparing the block check codes. The selection and repair module includes: a peak noise unit and a dynamic selection unit; The peak noise unit is used to calculate the local PSNR value of each color channel of the image, and the weighted average is used to obtain the peak signal-to-noise ratio of the image. The dynamic selection unit is used to detect network transmission status, including transmission speed, latency and jitter, calculate dynamic repair threshold, and determine the image frames that need to be repaired.

8. The intelligent image recognition and restoration processing system based on big data according to claim 7, characterized in that: The grayscale restoration module includes: a boundary recognition unit, a gradient diffusion unit, and a grayscale restoration unit; The boundary recognition unit is used to generate a grayscale image using a weighted grayscale formula, detect weak edges using the Sobel operator, and detect a complete grayscale boundary mask. The gradient diffusion unit is used to generate iso-illuminance lines with a width of one pixel, and performs texture diffusion along the iso-illuminance line direction based on the Perona-Malik diffusion equation. The grayscale restoration unit is used to align the isoluminance lines of adjacent frames and restore the grayscale image to a colored image.

9. The intelligent image recognition and restoration processing system based on big data according to claim 8, characterized in that: The color separation enhancement module includes: a color separation unit and a correlation elimination unit; The color separation unit is used to determine different color separation schemes, including RGB, HSV and Lab color separation; The correlation elimination unit performs KL transform to decorrelate each color component of the image and calculates the information content for each component.

10. The intelligent image recognition and restoration processing system based on big data according to claim 9, characterized in that: The color mixing and filtering module includes: a selection filtering unit, a color mixing unit, and an information recovery unit; The selection filtering unit is used to dynamically select the filter of the corresponding depth according to the amount of color separation information to perform monochrome noise reduction operation; The color mixing unit is used to weight and fuse the color components of each color separation scheme according to the information entropy ratio to restore the original image; The information restoration unit is used to verify the information entropy of the original image and the restored image, and to fine-tune the image regions where there are differences.

Images