A method for contrast enhancement of pelvic floor rehabilitation effect images

Abstract

The invention discloses a method for enhancing the contrast of pelvic floor rehabilitation effect images, which belongs to the field of image processing; the method aims to improve the quality of pelvic floor X-ray images and the efficiency and accuracy of clinical diagnosis; the method realizes full-process image quality optimization through acquisition compression and enhancement and restoration modular design; the acquisition compression module retains low-contrast area information through dynamic resolution adjustment and region-aware compression to reduce transmission bandwidth requirements; the enhancement and restoration module uses hybrid super-resolution to restore image resolution and enhance local and global contrast; in addition, the method also includes degradation modeling, loss function and training strategy and system integration module, and optimizes the enhancement effect of generating a network on low-contrast images through multi-task learning training; the invention not only solves the shortcomings of image degradation and noise processing, but also optimizes the balance between image enhancement effect and diagnostic relevance by introducing a multi-task loss function, providing strong support for clinical diagnosis and treatment.

Description

Technical Field

[0001] The invention belongs to the field of image processing, and in particular relates to a method for enhancing the contrast of pelvic floor rehabilitation effect images. Background Art

[0002] Contrast enhancement of medical images plays an important role in disease diagnosis and treatment monitoring, especially in the field of pelvic floor rehabilitation. The evaluation of pelvic floor rehabilitation effects often relies on X-ray images. However, due to the low resolution, low contrast and noise artifacts of pelvic floor area images, it is difficult to accurately capture details during image interpretation. Traditional image processing methods such as histogram equalization and simple filtering can improve contrast to a certain extent, but often cannot guarantee detail clarity and global image consistency at the same time. Therefore, it is particularly necessary to develop a special image contrast enhancement method for the special imaging characteristics of pelvic floor rehabilitation effects. This method can not only improve image quality, but also improve the efficiency and accuracy of clinical diagnosis and provide clearer image support.

[0003] Existing medical image contrast enhancement technologies have multiple limitations. First, most methods cannot take into account both local and global contrast enhancement at the same time. In images of pelvic floor rehabilitation effects, low-contrast areas coexist with noise artifacts, which makes it difficult for simple global adjustment methods to meet the needs. Second, during the compression and transmission of images, due to data loss and degradation problems, the image quality further deteriorates, making it more difficult for doctors to identify key areas. In addition, most existing image enhancement algorithms are based on single-task optimization and fail to optimize image quality and diagnostic relevance in combination with specific clinical application scenarios. Therefore, a comprehensive technical framework is needed that can significantly improve the contrast and diagnostic quality of pelvic floor rehabilitation images throughout the entire process of compression, transmission, and recovery.

[0004] In view of the above problems, the present invention proposes an innovative pelvic floor rehabilitation effect image contrast enhancement method, which combines the modular design of acquisition compression and enhancement restoration, and realizes the full-process image quality optimization through dynamic adjustment of resolution, region-aware compression and hybrid super-resolution restoration technologies; the acquisition compression module focuses on the retention of low-contrast area information, and ensures the quality of transmitted images through strategies such as weighted complexity calculation and local attention weight allocation; the enhancement restoration module further restores and optimizes the global and local details of the image through nonlinear normalization, adaptive filtering and dynamic contrast enhancement; this method not only solves the shortcomings of traditional methods in image degradation and noise processing, but also optimizes the balance between image enhancement effect and diagnostic relevance of the generation network by introducing a multi-task loss function, significantly improves the image quality, and provides strong support for clinical diagnosis and treatment. Summary of the invention

[0005] The present invention provides a method for contrast enhancement of pelvic floor rehabilitation effect images, aiming to significantly improve the contrast and diagnostic quality of pelvic floor rehabilitation images, thereby providing clearer image support for clinical diagnosis and treatment.

[0006] The present invention proposes a method for enhancing the contrast of pelvic floor rehabilitation effect images, which specifically comprises the following steps:

[0007] S1, collect and produce pelvic floor X-ray data sets, and generate training data pairs containing low resolution, low contrast and noise artifacts;

[0008] S2, constructing an acquisition compression PCM module, dynamically adjusting the resolution and performing region-aware compression on the input medical images, giving priority to retaining low-contrast region information, and generating low-resolution compressed images;

[0009] S3, build a degradation modeling DGM module to simulate the contrast degradation process in image transmission and build a degradation model of low-contrast artifacts and uneven illumination;

[0010] S4, construct the loss function and training strategy LTS module, define the comprehensive loss function and perform multi-task learning training to optimize the enhancement effect of the generated network on low-contrast images;

[0011] S5, construct an enhanced restoration DRM module, receive low-resolution compressed images, use a hybrid super-resolution generation network to restore image resolution, and enhance local and global contrast;

[0012] S6. Build a module integration system that integrates acquisition and enhanced processing modules to complete the full process of image compression, transmission, restoration and contrast enhancement.

[0013] Preferably, in step S2, constructing the acquisition compression PCM module specifically includes the following steps:

[0014] Step S21: Calculate the weighted complexity of the input image , and then calculate the scaling factor S to achieve dynamic resolution compression, so as to give priority to retaining the spatial information of low-contrast areas. Through dynamic resolution adjustment and region-aware compression, low-contrast area information is given priority to generate low-resolution compressed images. This function can significantly reduce the bandwidth requirements for data transmission, while ensuring the retention of key medical information and improving data processing efficiency. The specific calculation formula is:

[0015] ;

[0016] ;

[0017] Where S represents the resolution scaling factor, and its value range is (0,1]; Represents the weighted complexity of the input image; is the gradient weight of pixel (x, y), which is used to highlight the information of areas with lower contrast; is the intensity value of the pixel (x, y) in the input image; H and W are the height and width of the input image respectively.

[0018] Step S22: Based on the weighted edge strength of the image , dividing the image into key areas and non-critical areas , and perform region-aware compression on both, distinguishing key areas from non-key areas according to the weighted edge strength of the image, and using different compression strategies for different areas; this method effectively reduces unnecessary redundant data and enhances the transmission accuracy of key medical information; the specific calculation formula is:

[0019] ;

[0020] ;

[0021] in, Represents the compressed image; is a nonlinear compression operator; and A higher compression quality factor for critical areas and a lower compression quality factor for non-critical areas, respectively; and They represent the key areas with higher weighted edge strength and the non-key areas with lower weighted edge strength respectively.

[0022] Step S23: Calculate local attention weight , the specific calculation formula is:

[0023] ;

[0024] in, is the local attention weight, which is used to highlight contrast and structural features; It is the noise reduction intensity control parameter; is the edge weight adjustment factor; Represents the gradient strength of pixel p.

[0025] Step S24, the image is subjected to noise smoothing processing by using the local attention weight to obtain the PCM module compression output, the characteristics of the low contrast area are highlighted by using the local attention weight, the image is subjected to noise smoothing processing, the image transmission quality is ensured, and the visualization effect of the contrast and structural features is improved; the specific calculation formula is:

[0026] ;

[0027] in, Represents the value of the smoothed image at pixel (x, y); Represents the local neighborhood of pixel (x,y).

[0028] Preferably, in step S3, constructing the degradation modeling module DGM module specifically includes the following steps:

[0029] Step S31, using the optimized formula to simulate the contrast degradation caused by ringing artifacts caused by uneven X-ray exposure, the specific calculation formula is: providing real scene training data for the model, further optimizing the robustness of the enhancement algorithm; the specific calculation formula is:

[0030] ;

[0031] ;

[0032] in, represents the input image pixel; Indicates contrast degradation caused by uneven illumination; Indicates contrast degradation caused by ringing artifacts; is the frequency parameter; The intensity of X-ray illumination varies.

[0033] Step S32: Add degradation factors to generate blur and noise through the optimized formula to obtain the final degraded image. By adding blur, downsampling and noise interference, a degraded image containing multiple defects is generated to provide challenging training samples for the subsequent enhancement module to ensure the adaptability of the algorithm in complex scenes. The specific calculation formula is:

[0034] ;

[0035] in, is the original X-ray image; g is the Gaussian blur kernel; It is a downsampling operation; For noise interference; Indicates contrast degradation caused by uneven illumination; Indicates contrast degradation caused by ringing artifacts.

[0036] Preferably, in step S4, constructing the loss function and training strategy LTS module specifically includes the following steps:

[0037] Step S41, define a comprehensive loss function for optimizing the contrast enhancement and edge detail restoration of the generated network. The generated network is optimized by the comprehensive loss function, focusing on contrast enhancement, structure preservation, and edge detail restoration. This function significantly improves the overall performance of image enhancement. The specific calculation formula is:

[0038] ;

[0039] Where, L is the total comprehensive loss; is the contrast perception loss; To maintain losses for the structure; Enhance loss for edges; is the weight coefficient of each loss item in the overall loss;

[0040] Step S42: Design a contrast perception loss function based on the application scenario , the specific calculation formula is:

[0041] ;

[0042] in, and The grayscale value of the generated image and the real image at the pixel position (x, y); and is the horizontal and vertical gradient of the real image at the pixel position (x, y); the numerator of the formula as a whole represents the absolute error in pixel values ​​between the generated image and the real image; the denominator of the formula as a whole represents the loss weight for adjusting the low contrast area, and the cross gradient term is used to strengthen the attention of the area with weak contrast.

[0043] Step S43: Design a structure-preserving loss function according to the application scenario , the specific calculation formula is:

[0044] ;

[0045] in, To prevent a tiny positive number with a denominator of zero; the numerator of the formula as a whole represents the dot product of the generated image and the real image, which is used to measure the correlation between pixels; the denominator of the formula as a whole represents the normalization term, which calculates the norm product of the generated image and the real image to ensure that the correlation value is in the range of [0,1].

[0046] Step S44: Design edge enhancement loss function according to application scenario , the specific calculation formula is:

[0047] ;

[0048] in, and Represents the second-order gradient of the generated image and the real image in the horizontal direction; and Represents the second-order gradient of the generated image and the real image in the vertical direction; and Represents the second-order gradient difference between the generated image and the real image edge, which is used to optimize the edge clarity.

[0049] Step S45: During the training process, combine the segmentation task Optimize the diagnostic relevance of the generated network, the total loss function The specific calculation formula is:

[0050] ;

[0051] in, is the segmentation loss based on the Dice coefficient, is the weight coefficient of the segmentation loss, which is used to control the impact of the segmentation task on the overall optimization.

[0052] Preferably, in step S5, constructing the enhanced recovery DRM module specifically includes the following steps:

[0053] Step S51: receiving the low-resolution compressed image collected and transmitted , and adjust the pixel value range through normalization processing through improved nonlinear normalization function. After receiving the low-resolution compressed image, the pixel value is adjusted to a reasonable range through nonlinear normalization, and the resolution of the image is restored by using the hybrid super-resolution generation network; it can reconstruct the lost detail information, make the image more refined, and facilitate analysis; the specific calculation formula is:

[0054] ;

[0055] Among them, k is the stretching parameter that controls the nonlinear compression of the pixel value range.

[0056] Step S52: local detail enhancement is performed on the normalized image through direction-weighted adaptive filtering, and details in different directions are enhanced to restore and highlight subtle edge and texture features; the restoration of image details is improved, providing doctors with clearer observation conditions; the specific calculation formula is:

[0057] ;

[0058] in, is the enhanced local detail image; M is the number of directional filters; is the direction weight; To enhance the strength coefficient; is the convolution operation; is the filter in the mth direction; tanh() represents the hyperbolic tangent function.

[0059] Step S53, the image is globally enhanced by dynamically adjusting the contrast of the nonlinear mapping function, which can optimize the brightness level and contrast effect of the image and make the diagnosis area more prominent; the specific calculation formula is:

[0060] ;

[0061] in, This is the image after global contrast enhancement; is the local brightness mean of the image; Enhanced strength control parameters; Adjust the index for contrast.

[0062] Preferably, in step S6, a module integration system is constructed, specifically, firstly performing PCM module image dynamic compression on batch training images, then performing DGM module image degradation simulation, and then inputting them into the LTS module to adjust the comprehensive loss, then generating a network model through multi-task learning training and passing the generated network model into the DRM module, and finally in the application stage inputting the real image to be processed into the DRM module for image contrast enhancement to obtain the final image enhancement output; the integrated system realizes full-process automated processing from acquisition, compression, transmission to recovery and enhancement. The system reduces the need for manual intervention, improves the overall processing efficiency, and ensures that the image quality meets the diagnostic requirements.

[0063] Compared with the existing technology, the present invention realizes the optimization processing of the whole process from image acquisition to compression, transmission and recovery through modular design; the acquisition module pays special attention to the retention of information in low-contrast areas, and effectively reduces the data transmission bandwidth demand through weighted complexity calculation and local attention weight allocation, while ensuring the retention of key medical information; the enhancement and restoration module uses a hybrid super-resolution generation network to restore image resolution and enhance local and global contrast, significantly improving image quality; in addition, the present invention introduces a multi-task loss function to optimize the balance of the generation network on image enhancement effect and diagnostic relevance, providing stronger support for clinical diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 The figure is an overall flow chart of a method for contrast enhancement of pelvic floor rehabilitation effect images.

[0065] Figure 2 This is a schematic diagram of the structure flow of the acquisition and compression PCM module.

[0066] Figure 3 Schematic diagram of the DGM module structure flow for degradation modeling.

[0067] Figure 4 This is a schematic diagram of the enhanced recovery DRM module structure flow.

[0068] Figure 5 The figure is a schematic diagram of the overall structural flow of the integrated system.

[0069] Figure 6 Schematic diagram of image comparison before and after enhancement processing. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0071] Please see attached Figure 1 -Attached Figure 6 , the present invention provides a method for enhancing the contrast of pelvic floor rehabilitation effect images.

[0072] As attached Figure 1 The present invention proposes a method for enhancing the contrast of pelvic floor rehabilitation effect images, and the specific implementation method includes the following steps:

[0073] S1, collect and produce pelvic floor X-ray data sets, and generate training data pairs containing low resolution, low contrast and noise artifacts;

[0074] S2, constructing an acquisition compression PCM module, dynamically adjusting the resolution and performing region-aware compression on the input medical images, giving priority to retaining low-contrast region information, and generating low-resolution compressed images;

[0075] S3, build a degradation modeling DGM module to simulate the contrast degradation process in image transmission and build a degradation model of low-contrast artifacts and uneven illumination;

[0076] S4, construct the loss function and training strategy LTS module, define the comprehensive loss function and perform multi-task learning training to optimize the enhancement effect of the generated network on low-contrast images;

[0077] S5, construct an enhanced restoration DRM module, receive low-resolution compressed images, use a hybrid super-resolution generation network to restore image resolution, and enhance local and global contrast;

[0078] S6. Build a module integration system that integrates acquisition and enhanced processing modules to complete the full process of image compression, transmission, restoration and contrast enhancement.

[0079] Further, as attached Figure 1The acquisition and production of the pelvic floor X-ray dataset described in S1 is specifically to use modern medical imaging equipment to acquire pelvic floor X-ray images and generate a training dataset containing low resolution, low contrast and noise artifacts; the dataset format adopts the standard DICOM format, and the images are normalized through preprocessing so that the image data dimension is 512×512 pixels.

[0080] Further, as attached Figure 1 The construction steps of the acquisition compression PCM module described in S2 are as follows: Figure 2 As shown, the gradient weight is set to 0.1, the scaling factor range is 0.5-1.0, and the interpolation method is the cv2.INTER_AREA method in the opencv library; the specific implementation of the module includes the following steps:

[0081] Furthermore, step S21 calculates the weighted complexity of the input image , and then calculate the scaling factor S to achieve dynamic resolution compression, so as to give priority to retaining the spatial information of low-contrast areas. Through dynamic resolution adjustment and region-aware compression, low-contrast area information is given priority to generate low-resolution compressed images. This function can significantly reduce the bandwidth requirements for data transmission, while ensuring the retention of key medical information and improving data processing efficiency. The specific calculation formula is:

[0082] ;

[0083] ;

[0084] Where S represents the resolution scaling factor, and its value range is (0,1]; Represents the weighted complexity of the input image; is the gradient weight of pixel (x, y), which is used to highlight the information of areas with lower contrast; is the intensity value of the pixel (x, y) in the input image; H and W are the height and width of the input image respectively.

[0085] Further, step S22 is based on the weighted edge strength of the image , dividing the image into key areas and non-critical areas , and perform differential compression on the two. According to the weighted edge strength of the image, the key area and non-key area are distinguished, and different compression strategies are adopted for different areas. This method effectively reduces unnecessary redundant data and enhances the transmission accuracy of key medical information. The specific calculation formula is:

[0086] ;

[0087] ;

[0088] in, Represents the compressed image; is a nonlinear compression operator; and A higher compression quality factor for critical areas and a lower compression quality factor for non-critical areas, respectively; and They represent the key areas with higher weighted edge strength and the non-key areas with lower weighted edge strength respectively.

[0089] Further, step S23 calculates the local attention weight , the specific calculation formula is:

[0090] ;

[0091] in, is the local attention weight, which is used to highlight contrast and structural features; It is the noise reduction intensity control parameter; is the edge weight adjustment factor; Represents the gradient strength of pixel p.

[0092] Furthermore, step S24 performs noise smoothing on the image by using the local attention weight to obtain the PCM module compression output, highlights the features of the low contrast area by using the local attention weight, performs noise smoothing on the image, ensures the image transmission quality, and improves the visualization effect of contrast and structural features; the specific calculation formula is:

[0093] ;

[0094] in, Represents the value of the smoothed image at pixel (x, y); Represents the local neighborhood of pixel (x,y).

[0095] Further, as attached Figure 1 The degradation modeling module DGM module described in S3 is constructed. The specific construction process of the module is as shown in the attached Figure 3 As shown, the Gaussian blur kernel is set to 5×5, the blur intensity σ is 1.5, and the noise level is 0.05; the specific implementation of the module includes the following steps:

[0096] Furthermore, step S31 simulates the contrast degradation caused by ringing artifacts caused by uneven X-ray exposure through the optimized formula calculation, providing real scene training data for the model and further optimizing the robustness of the enhancement algorithm; the specific calculation formula is:

[0097] ;

[0098] ;

[0099] in, represents the input image pixel; Indicates contrast degradation caused by uneven illumination; Indicates contrast degradation caused by ringing artifacts; is the frequency parameter; The intensity of X-ray illumination varies.

[0100] Furthermore, step S32 adds degradation factors to generate blur and noise through the optimized formula to obtain the final degraded image. By adding blur, downsampling and noise interference, a degraded image containing multiple defects is generated to provide challenging training samples for the subsequent enhancement module to ensure the adaptability of the algorithm in complex scenes. The specific calculation formula is:

[0101] ;

[0102] in, is the original X-ray image; g is the Gaussian blur kernel; It is a downsampling operation; For noise interference; Indicates contrast degradation caused by uneven illumination; Indicates contrast degradation caused by ringing artifacts.

[0103] Further, as attached Figure 1 The loss function and training strategy LTS module described in S4 is constructed, wherein the contrast loss weight is set to 0.5; the structure loss weight is set to 0.3; and the edge loss weight is set to 0.2. The specific implementation of the module includes the following steps:

[0104] Further, in step S41, a comprehensive loss function is defined to optimize the contrast enhancement and edge detail restoration of the generated network. The generated network is optimized by the comprehensive loss function, focusing on contrast enhancement, structure preservation, and edge detail restoration. This function significantly improves the overall performance of image enhancement. The specific calculation formula is:

[0105] ;

[0106] Where, L is the total comprehensive loss; is the contrast perception loss; To maintain losses for the structure; Enhance loss for edges; is the weight coefficient of each loss item in the overall loss.

[0107] Further, step S42 specifically designs a contrast perception loss function according to the application scenario , the specific calculation formula is:

[0108] ;

[0109] in, and The grayscale value of the generated image and the real image at the pixel position (x, y); and is the horizontal and vertical gradient of the real image at the pixel position (x, y); the numerator of the formula as a whole represents the absolute error in pixel values ​​between the generated image and the real image; the denominator of the formula as a whole represents the loss weight for adjusting the low contrast area, and the cross gradient term is used to strengthen the attention of the area with weak contrast.

[0110] Further, step S43 specifically designs the structure preservation loss function according to the application scenario , the specific calculation formula is:

[0111] ;

[0112] in, To prevent a tiny positive number with a denominator of zero; the numerator of the formula as a whole represents the dot product of the generated image and the real image, which is used to measure the correlation between pixels; the denominator of the formula as a whole represents the normalization term, which calculates the norm product of the generated image and the real image to ensure that the correlation value is in the range of [0,1].

[0113] Further, step S44 specifically designs the edge enhancement loss function according to the application scenario , the specific calculation formula is:

[0114] ;

[0115] in, and Represents the second-order gradient of the generated image and the real image in the horizontal direction; and Represents the second-order gradient of the generated image and the real image in the vertical direction; and Represents the second-order gradient difference between the generated image and the real image edge, which is used to optimize the edge clarity.

[0116] Furthermore, step S45 combines the segmentation task during the training process. Optimize the diagnostic relevance of the generated network, the total loss function The specific calculation formula is:

[0117] ;

[0118] in, is the segmentation loss based on the Dice coefficient, is the weight coefficient of the segmentation loss, which is used to control the impact of the segmentation task on the overall optimization.

[0119] Further, as attached Figure 1 The enhanced recovery DRM module described in S5 is constructed, and the specific structure of the DRM module is as shown in the attached Figure 4 As shown; the horizontal and vertical direction weights are set to 0.25; the enhancement factor is 1.5; the filter dimension is 3×3; the specific implementation of the module includes the following steps:

[0120] Further, step S51 receives the low-resolution compressed image collected and transmitted , and adjust the pixel value range through normalization processing through improved nonlinear normalization function. After receiving low-resolution compressed images, the pixel values ​​are adjusted to a reasonable range through nonlinear normalization, and the resolution of the image is restored using the hybrid super-resolution generation network; it can reconstruct the lost detail information, make the image more refined, and facilitate the doctor's analysis; the specific calculation formula is:

[0121] ;

[0122] Among them, k is the stretching parameter that controls the nonlinear compression of the pixel value range.

[0123] Furthermore, step S52 performs detail enhancement on the normalized image through directional weighted adaptive filtering, performs local detail enhancement on details in different directions, restores and highlights subtle edge and texture features, improves the restoration of image details, and provides doctors with clearer observation conditions. The specific calculation formula is:

[0124] ;

[0125] in, is the enhanced local detail image; M is the number of directional filters; is the direction weight; To enhance the strength coefficient; is the convolution operation; is the filter in the mth direction; tanh() represents the hyperbolic tangent function.

[0126] Furthermore, step S53 performs global contrast enhancement on the image by dynamically adjusting the contrast of the nonlinear mapping function, which can optimize the brightness level and contrast effect of the image and make the diagnosis area more prominent; the specific calculation formula is:

[0127] ;

[0128] in, This is the image after global contrast enhancement; is the local brightness mean of the image; Enhanced strength control parameters; Adjust the index for contrast.

[0129] Further, as attached Figure 1 The building block integration system described in S6 is shown in the attached figure. Figure 5 As shown in the figure; the specific implementation process is to first perform dynamic compression of the batch training images in the PCM module, then perform image degradation simulation in the DGM module, and then input them into the LTS module to adjust the comprehensive loss. Subsequently, a network model is generated through multi-task learning training and the generated network model is passed to the DRM module. Finally, in the application stage, the real image to be processed is input into the DRM module for image contrast enhancement to obtain the final image enhancement output; the integrated system realizes the full process automation processing from acquisition, compression, transmission to recovery and enhancement. The overall design process of the system is shown in the attached figure. Figure 5 The system reduces the need for manual intervention, improves overall processing efficiency, and ensures that the image quality meets diagnostic requirements. The image comparison diagram before and after processing by the system is shown in the attached figure. Figure 6 As shown, part a is the original image before processing, and part b is the output image after processing and enhancement.

Claims

1. A method for enhancing contrast of pelvic floor rehabilitation effect images, characterized in that: The following steps are involved: S1, collect and produce pelvic floor X-ray data sets, and generate training data pairs containing low resolution, low contrast and noise artifacts; S2, constructing an acquisition compression PCM module, performing dynamic resolution compression adjustment and region-aware compression on the image data set collected by S1, giving priority to retaining low-contrast area information, and generating a first-stage low-resolution compressed image; constructing an acquisition compression PCM module includes the following steps: S21, by calculating the weighted complexity W of the input image C (I), and then calculate the scaling factor S to achieve dynamic resolution compression adjustment to prioritize the spatial information of low-contrast areas. The specific calculation formula is: Where S represents the resolution scaling factor, and its value range is (0, 1]; W C (I) represents the weighted complexity of the input image; ω(x, y) is the gradient weight of pixel (x, y), which is used to highlight the information of the area with lower contrast; I(x, y) is the intensity value of pixel (x, y) in the input image; H and W are the height and width of the input image respectively; S22, according to the weighted edge strength E of the image weighted (I) The image after dynamic resolution compression is divided into key areas R key and non-critical area R non-key , and perform region-aware compression on both. The specific calculation formula is: Among them, I compressed Indicates the compressed image; Q q (I) is a nonlinear compression operator; q high and q low A higher compression quality factor is used for critical areas and a lower compression quality factor is used for non-critical areas; R key and R non-key They represent the key areas with higher weighted edge strength and the non-key areas with lower weighted edge strength respectively; S23. Calculate the local attention weight A(x, y, p). The specific calculation formula is: Among them, A(x, y, p) is the local attention weight, which is used to highlight the contrast and structural features; σ is the noise reduction intensity control parameter; α is the edge weight adjustment factor; represents the gradient strength of pixel p; S24, performing noise smoothing on the image after regional perception compression by using the local attention weight to obtain the PCM module compression output, and the specific calculation formula is: Among them, I denoised (x, y) represents the value of the smoothed image at pixel (x, y); Ω represents the local neighborhood of pixel (x, y); S3, constructing a degradation modeling DGM module, simulating the contrast degradation process in image transmission, constructing a degradation model of low-contrast artifacts and uneven illumination, degrading the first-stage low-resolution compressed image, and obtaining a final degraded image; constructing the degradation modeling DGM module includes the following steps: S31. Calculate using the optimized formula to simulate the contrast degradation caused by ringing artifacts caused by uneven X-ray exposure. The specific calculation formula is: FLUENT illum (x,y)=I p (x,y)·(1+β·sin(2πfx))? A ring (x,y)=I p (x,y)*Sinc(x,y); Among them, I p (x, y) is the pixel of the low-resolution compressed image obtained in step S2; A illum (x, y) represents the contrast degradation caused by uneven illumination; A ring (x, y) represents the contrast degradation caused by ringing artifacts; f is the frequency parameter; β is the intensity of X-ray illumination variation; S32, adding degradation factors to generate blur and noise through the optimized formula to obtain the final degraded image. The specific calculation formula is: Among them, I q is the low-resolution compressed image obtained in step S2; g is the Gaussian blur kernel; ↓s is the downsampling operation; η is the noise interference; A illum (x, y) represents the contrast degradation caused by uneven illumination; A ring (x, y) represents the contrast degradation caused by ringing artifacts; S4, constructing a loss function and training strategy LTS module, defining a comprehensive loss function and performing multi-task learning training, and optimizing the enhancement effect of the hybrid super-resolution generation network on low-contrast images; constructing a loss function and training strategy LTS module includes the following steps: S41. Define a comprehensive loss function to optimize the contrast enhancement and edge detail restoration of the hybrid super-resolution generation network. The specific calculation formula is: L=λ contrast L contrast +λ structure L structure +λ edge L edge ; Where, L is the total comprehensive loss; L contrast is the contrast perception loss; L structure To maintain the loss of structure; L edge is the edge enhancement loss; contrast ,λ structure ,λ edge is the weight coefficient of each loss item in the overall loss; S42. Design the contrast perception loss function L according to the application scenario contrast , the specific calculation formula is: Among them, I gen (x, y) and I real (x, y) is the grayscale value of the generated image and the real image at the pixel position (x, y); and is the horizontal and vertical gradient of the real image at the pixel position (x, y); the numerator of the formula as a whole represents the absolute error in pixel values ​​between the generated image and the real image; the denominator of the formula as a whole represents the loss weight for adjusting the low-contrast area, and the cross-gradient term is used to strengthen the focus of the area with weak contrast; S43. Design the structure preservation loss function L according to the application scenario structure , the specific calculation formula is: Among them, ∈ is a small positive number to prevent the denominator from being zero; the numerator of the formula as a whole represents the dot product of the generated image and the real image, which is used to measure the correlation between pixels; the denominator of the formula as a whole represents the normalization term, which calculates the norm product of the generated image and the real image to ensure that the correlation value is in the range of [0, 1]; S44. Design edge enhancement loss function L according to application scenario edge , the specific calculation formula is: in, and Represents the second-order gradient of the generated image and the real image in the horizontal direction; and Represents the second-order gradient of the generated image and the real image in the vertical direction; and Represents the second-order gradient difference between the generated image and the real image edge, which is used to optimize the edge clarity; S45. During the training process, combined with the segmentation task L seg Optimizing the diagnostic relevance of the hybrid super-resolution generative network, the total loss function L total The specific calculation formula is: L total =L+λ seg L seg ; Among them, L seg is the segmentation loss based on the Dice coefficient, λ seg is the weight coefficient of the segmentation loss, which is used to control the impact of the segmentation task on the overall optimization; S5, construct an enhanced restoration DRM module, receive the real image to be processed, use the hybrid super-resolution generation network to restore the image resolution, and enhance the local and global contrast; S6. Build a module integration system that integrates acquisition and enhanced processing modules to complete the full process of image compression, transmission, restoration and contrast enhancement.

Images