Image processing method and system for enhanced skin pressure injury monitoring

Abstract

The invention discloses an image processing method and system for enhanced skin pressure injury monitoring, and relates to the technical field of image processing, and the method comprises the steps: carrying out the adaptive partitioning processing of a skin image; extracting texture entropy, gradient distribution and brightness deviation characteristics of each local block, and constructing a local micro-damage response diagram; constructing an adaptive enhancement function and performing enhancement processing; executing interactive verification of the multiple sets of enhanced images, calculating a consistency score and executing interactive enhancement processing; executing offset residual analysis to construct a local enhancement credibility graph and triggering secondary adaptive enhancement processing; and performing interactive fusion on the first enhanced image and the second enhanced image, and outputting a third enhanced image. The technical problems that in the prior art, conventional image enhancement cannot adapt to local heterogeneity of skin images, early weak injury features are difficult to effectively extract, and the credibility is low are solved, and the technical effect that accurate enhancement processing of the local micro injury features is achieved to ensure that subsequent image labeling results are reliable is achieved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an image processing method and system for enhanced monitoring of skin pressure injuries. Background Technology

[0002] Traditional monitoring of pressure injuries relies on visual examination and manual palpation by healthcare professionals, which suffers from high subjectivity, difficulty in identifying minute early-stage injuries, and low efficiency of manual monitoring. Skin surface images are increasingly becoming an important basis for auxiliary diagnosis. However, skin images are often affected by factors such as uneven lighting, differences in shooting angles, natural variations in skin texture, and low contrast in early-stage injury areas during actual acquisition, making it difficult to accurately detect minute or early-stage pressure injuries. Existing image processing techniques, such as histogram equalization or contrast stretching enhancement strategies, can improve overall image visibility. However, when processing local weak features such as early-stage pressure injuries, global enhancement may amplify irrelevant noise or normal skin texture, interfering with the identification of the true injury area. Skin in different locations exhibits high heterogeneity due to pressure levels, subcutaneous structures, and lighting conditions. A single enhancement function is insufficient to adapt to the characteristics of each region. The differences between early-stage injury areas and surrounding healthy tissue are very subtle, and conventional enhancement can easily cause feature overload or distortion. Furthermore, local enhancement based on region segmentation often uses non-overlapping partitions, which can easily lead to enhancement discontinuities at block boundaries and may result in incomplete coverage of minute injuries, making it difficult to ensure the stability and reliability of enhancement under complex conditions.

[0003] Therefore, current related technologies suffer from technical problems such as conventional image enhancement being unable to adapt to the local heterogeneity of skin images, difficulty in effectively extracting early and weak damage features, and low reliability. Summary of the Invention

[0004] This application provides an image processing method and system for enhanced monitoring of skin pressure injuries, which solves the technical problems in the prior art where conventional image enhancement cannot adapt to the local heterogeneity of skin images, is difficult to effectively extract early weak damage features, and has low reliability. It achieves the technical effect of accurately enhancing local micro-damage features to ensure the reliability of subsequent image annotation results.

[0005] This application provides an image processing method for enhanced monitoring of skin pressure injuries. The method includes: adaptively partitioning a skin image to be processed to establish N local blocks, wherein the N local blocks include repeated partitioning blocks; extracting texture entropy, gradient distribution, and brightness deviation features of each local block, and constructing a local micro-damage response map based on the extraction results; constructing an adaptive enhancement function for each local block based on the local micro-damage response map, and applying different adaptive enhancement functions to the repeated partitioning blocks of the same original skin image to construct multiple sets of enhanced images; performing interactive verification of the multiple sets of enhanced images, calculating the consistency score of the local micro-damage response corresponding to each repeated enhanced image, and performing interactive enhancement processing based on the calculation results to establish a first enhanced image; performing offset residual analysis on the first enhanced image to construct a local enhancement confidence map, triggering a second adaptive enhancement process based on the local enhancement confidence map to establish a second enhanced image; and interactively fusing the first enhanced image and the second enhanced image to output a third enhanced image.

[0006] In a possible implementation, performing offset residual analysis on the first enhanced image to construct a local enhancement confidence map includes: calculating the response offset residual of the same local region under different enhancement paths for the enhancement results generated corresponding to different repeated segmentation blocks in the first enhanced image, under the condition of spatial alignment; using the response offset residual, calculating the local texture entropy offset, gradient direction offset, and brightness enhancement amplitude offset respectively, and combining the calculation results to generate a local enhancement stability index; using the local enhancement stability index to classify the confidence of the local region to form the local enhancement confidence map.

[0007] In a possible implementation, using the local enhancement stability index to classify the credibility of local regions further includes: after reading the current partitioning granularity, performing skin image repartitioning based on the correspondence between the partitioning granularity and the overall structural scale of the skin image to establish macro-partitions, wherein the area of ​​the macro-partitions is larger than the area of ​​the local partitions; based on the macro-partitions, performing intra-regional continuity and response consistency verification on the enhancement results within each macro-partition to establish a macro-response feature set characterizing enhancement stability at the macro scale; and performing joint mapping analysis between the macro-response feature set and the local enhancement stability index to complete the credibility classification.

[0008] In a possible implementation, a second enhanced image is constructed by triggering a secondary adaptive enhancement process based on the local enhancement confidence map. This includes: dividing local regions in the first enhanced image into confidence states according to the local enhancement confidence map, and constructing high-confidence enhancement region identifiers, medium-confidence enhancement region identifiers, and low-confidence enhancement region identifiers; constructing constraints to maintain the continuity of the original skin texture structure for the low-confidence enhancement regions, and limiting the brightness stretching amplitude and gradient enhancement intensity, and performing directional enhancement processing of pressure damage-related response features; reading the enhancement results of the corresponding regions in the first enhanced image in the medium-confidence enhancement regions, and performing local fine-tuning processing based on the reading results; extracting enhancement parameter distribution features for the high-confidence enhancement regions, establishing enhancement reference constraints, and using the enhancement reference constraints to guide the secondary adaptive enhancement process of the medium-confidence and low-confidence enhancement regions to construct the second enhanced image.

[0009] In a possible implementation, the first enhanced image and the second enhanced image are interactively fused to output a third enhanced image, including: under the condition of spatial alignment, calculating the enhancement offset in the dimensions of local texture, gradient response, and brightness response based on the first enhanced image and the second enhanced image, and constructing a local enhancement offset field characterizing the second enhanced image relative to the first enhanced image; performing confidence modulation processing on the local enhancement offset field based on the local enhancement confidence map to form a confidence-constrained enhancement offset modulation field; using the first enhanced image as the enhancement master state structure, performing controlled injection processing on the enhancement offset modulation field to introduce pressure damage-related correction enhancement components from the second enhanced image; and reconstructing the third enhanced image based on the controlled injection processing result.

[0010] In a possible implementation, flexible response parameter optimization is performed based on the flexible response function to output flexible control parameters. The method includes: the repeated segmentation and slicing is generated by applying different slicing strategies to the same skin image. The different slicing strategies include different slicing sizes, different slicing start offset positions, or different slicing shapes, so that the same image region is covered multiple times under different slicing strategies to construct repeated segmentation and slicing with spatial overlap.

[0011] In a possible implementation, the image processing method for enhanced skin pressure injury monitoring further includes: when constructing a local enhanced confidence map, performing spatial continuity constraint processing on the confidence grading results of adjacent local regions to suppress confidence abrupt changes caused by local noise or partition boundary effects.

[0012] This application also provides an image processing system for enhanced monitoring of skin pressure injuries. The system includes: an image partitioning module for adaptively partitioning a skin image to be processed, establishing N local blocks, wherein the N local blocks include repeated partitioning blocks; a block feature extraction module for extracting texture entropy, gradient distribution, and brightness deviation features of each local block, and constructing a local micro-damage response map based on the extraction results; and an image enhancement processing module for constructing an adaptive enhancement function for each local block based on the local micro-damage response map, and applying different adaptive enhancement functions to repeated partitioning blocks of the same original skin image. The system performs enhancement processing to construct multiple sets of enhanced images. A first enhanced image creation module is used in conjunction with the interactive enhancement processing module to perform interactive verification of the multiple enhanced images, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, and perform interactive enhancement processing based on the calculation results to create the first enhanced image. A second enhanced image creation module performs offset residual analysis on the first enhanced image, constructs a local enhancement confidence map, and triggers secondary adaptive enhancement processing based on the local enhancement confidence map to create the second enhanced image. A third enhanced image output module performs interactive fusion of the first and second enhanced images to output the third enhanced image.

[0013] This application proposes an image processing method and system for enhanced skin pressure injury monitoring. The method involves adaptively partitioning skin images, extracting texture entropy, gradient distribution, and brightness deviation features from each local block to construct a local micro-damage response map, constructing an adaptive enhancement function, performing enhancement processing, conducting interactive verification of multiple enhanced images, calculating consistency scores, performing interactive enhancement processing, performing offset residual analysis to construct a local enhancement reliability map, and triggering secondary adaptive enhancement processing, and interactively fusing the first and second enhanced images to output a third enhanced image. This addresses the technical problems of conventional image enhancement in existing technologies, such as inability to adapt to the local heterogeneity of skin images, difficulty in effectively extracting early weak damage features, and low reliability. It achieves precise enhancement processing of local micro-damage features to ensure reliable subsequent image annotation results. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1A schematic flowchart of the image processing method for enhanced skin pressure injury monitoring provided in the embodiments of this application.

[0016] Figure 2 A schematic diagram of the image processing system for enhanced skin pressure injury monitoring provided in this application embodiment.

[0017] Figure reference numerals: Image partitioning processing module 10, block feature extraction module 20, image enhancement processing module 30, first enhanced image creation module 40, second enhanced image creation module 50, third enhanced image output module 60. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0019] This application provides an image processing method for enhanced monitoring of skin pressure injuries, such as... Figure 1 As shown, the method includes: Step S100: Perform adaptive partitioning on the skin image to be processed to establish N local blocks, wherein the N local blocks include repeated partitioning blocks.

[0020] Preferably, the skin image to be processed undergoes adaptive partitioning, dividing the entire skin image into N local blocks, where N is a positive integer representing the number of local blocks. The blocks can be rectangular, square, or other geometric shapes, and their size, position, or shape can be adaptively adjusted according to the image content. For example, the block size can be dynamically determined based on skin texture density or illumination distribution. The N local blocks include overlapping partitioning, which means that different local blocks are allowed to have spatially overlapping areas during the partitioning process. For example, a fixed-size window, such as 64×64 pixels, slides across the skin image, with a sliding step size smaller than the window size, and adjacent windows have overlapping pixels. The system obtains repeated partitions by dividing the skin image into regions. This involves multi-scale overlapping partitioning, using blocks of different sizes such as 32×32, 64×64, and 128×128. These blocks spatially cover the same image region, resulting in repeated partitions. Alternatively, blocks of the same size can be divided using different starting coordinates such as (0,0) and (16,16) to form a network of overlapping, misaligned blocks, again yielding repeated partitions. This partitioning process ensures that the same pressure-related lesion feature is completely covered by multiple blocks. Furthermore, when the same pixel region appears in multiple blocks, features can be extracted based on different local contexts such as adjacent textures and gradient environments, enhancing the robustness of the analysis.

[0021] Furthermore, step S100 also includes the following: the repeated segmentation and slicing includes generating the same skin image by applying different segmentation strategies. The different segmentation strategies include different segmentation sizes, different segmentation start offset positions, or different segmentation shapes, so that the same image region is covered multiple times under different segmentation strategies, so as to construct repeated segmentation and slicing with spatial overlap.

[0022] Preferably, the same skin image is independently divided multiple times using different partitioning strategies to generate repeated partitioned blocks. These different partitioning strategies include different block sizes, different block start offsets, or different block shapes. The blocks generated by different partitioning strategies naturally have spatial overlap, ensuring that any region in the skin image to be processed is covered by multiple blocks from different sources. Specifically, different block sizes refer to using windows of various sizes to divide the skin image. For example, strategies A, B, and C use fixed sizes of 32×32, 64×64, and 128×128 pixels to divide the skin image. A 100×100 pixel region is covered by multiple small blocks from strategy A, a few medium-sized blocks from strategy B, and a single large block from strategy C. This region is analyzed multiple times in different scale contexts. Different block start offsets refer to using blocks of the same size starting from different points. Coordinate partitioning, for example, with a block size of 64×64 pixels, strategy 1 starts at offset (0,0), strategy 2 starts at offset (32,32), and strategy 3 starts at offset (16,48). The same point in the image may belong to the edge of a block in strategy 1, the center of another block in strategy 2, and a different position in a third block in strategy 3. This point is covered by three blocks with different spatial locations. Different block shapes refer to the use of non-rectangular block shapes for partitioning. For example, strategies α, β, and γ use circular, hexagonal, and irregular shapes based on superpixel segmentation, respectively. Irregular regions in the image containing early erythema may be cut by circular blocks, contained more completely by hexagonal blocks, and have their boundaries precisely fitted by superpixel blocks. This region is characterized multiple times under different geometric constraints. This ensures that each point in the image can be analyzed under multiple different local contexts and geometric constraints.

[0023] Step S200: Extract the texture entropy, gradient distribution, and brightness deviation features of each local block, and construct a local micro-damage response map based on the extraction results.

[0024] Preferably, the gray-level co-occurrence matrix or gray-level histogram of each local block is calculated, and the probability scalar value of gray level occurrence is calculated using the entropy formula to determine the texture entropy, which represents the degree of texture disorder in the block. In particular, early pressure damage may cause the local skin texture uniformity to be disrupted, resulting in subtle roughness or disorder, which manifests as an abnormal increase in texture entropy. Operators such as Sobel and Prewitt are used to calculate the gradient of each pixel in the x and y directions within the block, including statistically extracting the mean, variance, or histogram of the gradient magnitude of all pixels in the block, and simultaneously calculating the statistical distribution of the gradient direction angle to reflect the consistency of the gradient direction, thereby determining the gradient distribution, which represents the changes in edge intensity and direction within the block. In particular, the gradient distribution of normal skin texture is regular, while damaged areas may produce abnormal edge responses due to tissue swelling and discoloration, such as edge blurring leading to a decrease in gradient magnitude or inflammation spreading leading to gradient direction disorder.

[0025] Preferably, a brightness index is defined, such as the average gray value of the block or the average value of the L channel in the Lab color space. By calculating the difference between the brightness of the block and the average brightness of its directly adjacent peripheral area, or by calculating the Z-score of the brightness of the block and the average brightness of the skin area of ​​the whole image, a scalar value is output as a brightness deviation feature, indicating the degree of abnormality of the brightness of the block being too high or too low relative to the normal area. Among them, pressure damage is manifested as erythema, with local brightness reduced relative to the surrounding skin, or fading / ischemia, with local brightness increased.

[0026] Preferably, the Min-Max normalization function is used to normalize the texture entropy, gradient distribution, and brightness deviation features of each local block, mapping features of different dimensions to the same scale. Then, for each local block, the normalized texture entropy value, gradient distribution feature, and brightness deviation value are weighted and fused to calculate N micro-damage initial response values. The weight coefficients are set according to historical experimental data and feature importance. Then, each micro-damage initial response value is assigned to all pixels of the local block. Alternatively, each micro-damage initial response value is marked at the center of the local block. If the blocks overlap, the same pixel may obtain micro-damage initial response values ​​from multiple local blocks. Alternatively, for each pixel in the image, all local blocks containing the pixel are identified, and their corresponding micro-damage initial response values ​​are weighted and aggregated as the pixel's micro-damage response value. Finally, a local micro-damage response map of the same size as the original image is obtained. The higher the micro-damage response value of an element, the more the local area where the pixel is located matches the expected pattern of pressure damage in terms of texture, gradient, and brightness features, and the higher the probability of it being a suspected micro-damage area.

[0027] Step S300: Based on the local micro-damage response map, construct an adaptive enhancement function for each local block. For repeated divisions of the same original skin image, apply different adaptive enhancement functions to perform enhancement processing to construct multiple sets of enhanced images.

[0028] Preferably, based on the suspected damage level and regional characteristics represented by each local block in the local micro-damage response map, the most suitable adaptive enhancement function is constructed for that local block. Since there are repeated local blocks, the same image region may belong to multiple different local blocks. Therefore, each block has its own adaptive enhancement function. Different adaptive enhancement functions are applied to the skin image to be processed to perform different enhancement processes, and finally multiple sets of enhanced images are obtained. This avoids the drawbacks of insufficient enhancement of damaged areas and excessive enhancement of normal areas by global uniform enhancement, and achieves key enhancement of suspected areas while maintaining the original appearance of normal areas, thereby providing multi-view and multi-parameter enhancement performance of damaged areas.

[0029] Preferably, different adaptive enhancement functions are constructed for each local block, typically based on classic image enhancement operators, transforming the original pixel values ​​within the local block into enhanced pixel values. Specifically, the enhancement function may include a contrast stretching function, a gamma correction function, or a local histogram equalization function. The parameters of the adaptive enhancement function are dynamically adjusted according to the local micro-damage response characteristics of the local block. A higher average micro-damage response value within the local block indicates a greater likelihood of suspected damage, thus potentially resulting in a steeper contrast stretching slope and a gamma value deviating more from 1 to highlight details in dark or bright areas. Simultaneously, the dominant enhancement mode is determined based on the characteristics of the local block in the local micro-damage response map. If texture entropy contributes significantly, high-frequency enhancement is emphasized, or a filter emphasizing texture is used; if brightness deviation contributes significantly, brightness correction or chroma adjustment is emphasized; if gradient distribution contributes significantly, edge sharpening is emphasized. For local blocks with extremely low micro-damage response values, which may be normal skin, the enhancement function is set to an approximate identity transformation or only performs very slight smoothing to avoid over-processing and introducing noise.

[0030] Preferably, due to the existence of repeated partitioning and slicing, a pixel may belong to different local blocks. For a single local block, the corresponding original skin image region is extracted, and the response statistical features of the block region are extracted from the local micro-damage response map. An adaptive enhancement function is instantiated based on the response statistical features and applied to the image region to obtain the enhancement result of the local block. After independently enhancing all local blocks, the enhancement results corresponding to non-overlapping partitions are directly stitched together. For overlapping pixel regions, stitching is performed based on block affiliation or enhancement function weighting. Block affiliation means assigning a main block to each pixel, for example, assigning it to the largest block containing the pixel at its center and using the enhancement result of that main block. Enhancement function weighting means weighting the multiple possible enhancement results of a pixel, with the weights determined based on the spatial distance between each block and the pixel or the confidence level of the block's response value. For the same partitioning strategy, the enhancement results of all blocks are stitched together to determine a complete set of enhanced images. Finally, for repeated partitioning and slicing of the same original skin image, multiple sets of images with different enhancement perspectives are constructed.

[0031] Step S400: Perform interactive verification of multiple sets of enhanced images, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, perform interactive enhancement processing based on the calculation results, and establish the first enhanced image.

[0032] Preferably, interactive verification is performed on multiple sets of enhanced images to evaluate their reliability, select stable and reliable enhancement features, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, and generate a higher-quality first enhanced image through fusion optimization. Specifically, for each pixel or local region in each set of enhanced images, its micro-damage response value is recalculated based on texture entropy, gradient distribution, and brightness features. Since multiple sets of enhanced images originate from different repeated regions and blocks of the same original image and are spatially aligned, for each pixel position in the image, the inverse variance of its micro-damage response value in all sets of enhanced images is calculated as the consistency score. The closer the response values ​​of each enhancement result are and the smaller the variance, the higher the consistency score and the higher the reliability of the enhancement result at that point. Conversely, the lower the consistency score, the more discrepancies arise between different enhancement processes, and the enhancement result is unreliable or contains artifacts. Then, interactive enhancement processing is performed on multiple sets of enhanced images. This involves weighted fusion correction based on the calculated consistency scores. For regions with high consistency scores, the weighted average of the corresponding values ​​from each set of enhanced images is used to calculate the value of the first enhanced image at that location. The weights are proportional to the consistency score of that set of images at that point. The result for high-consistency regions is a smooth fusion of reliable enhancement results, suppressing random noise. For regions with low consistency scores, the inconsistency is determined to be due to noise, partition boundary effects, or weak features in the region leading to different sensitivities in different enhancement processes. Therefore, feature values ​​exhibiting moderate response in most enhanced images are selected to avoid being dominated by extreme values ​​of individual over-enhancement artifacts. Finally, reliable region information from one set of enhanced images is used to guide the correction of unreliable regions in another set of images, obtaining the first enhanced image. This preserves consistent and stable features across multiple enhancement results and intelligently arbitrates and smooths inconsistent image regions, resulting in overall reliability higher than any single set of enhanced images.

[0033] Step S500: Perform offset residual analysis on the first enhanced image, construct a local enhancement confidence map, trigger secondary adaptive enhancement processing based on the local enhancement confidence map, and establish a second enhanced image.

[0034] Step S500 further includes, for the enhancement results generated corresponding to different repeated segmentation blocks in the first enhanced image, under the condition of spatial alignment, calculating the response offset residual of the same local region under different enhancement paths; using the response offset residual, calculating the local texture entropy offset, gradient direction offset, and brightness enhancement amplitude offset respectively, and combining the calculation results to generate a local enhancement stability index; using the local enhancement stability index to classify the credibility of the local region to form the local enhancement credibility map.

[0035] Preferably, offset residual analysis is performed on the first enhanced image to quantify the stability of enhancement results from different sources within it, forming a local enhancement confidence map that reflects the reliability of the enhancement results. Specifically, for any pixel in the first enhanced image, all repeated partition blocks that have covered the pixel are identified, and intermediate enhancement results generated by different repeated partition blocks for the corresponding pixel are obtained. Under the condition of spatial alignment, the response offset residual of the same local region under different enhancement paths is calculated, that is, the dispersion between the response values ​​of multiple enhancement results at the pixel is calculated. Here, different enhancement paths represent different partitions. For example, the variance or standard deviation of feature quantities such as pixel intensity or gradient magnitude is calculated. The offset residual directly reflects the fluctuation of the output results when different enhancement paths process the same position. The greater the fluctuation, the more inconsistent or unstable the enhancement process is in processing the pixel.

[0036] Preferably, the local texture entropy offset, gradient direction offset, and brightness enhancement magnitude offset are calculated using the response offset residuals. The local texture entropy offset is determined by calculating the standard deviation or range of the local texture entropy values ​​of multiple intermediate enhancement results within a small window centered on the pixel. A larger local texture entropy offset indicates a greater difference in texture disorder assessment at that point due to different enhancement paths. Similarly, the gradient direction offset is determined by calculating the dominant gradient direction of multiple intermediate enhancement results within the small window at that point and calculating the circular square error or average direction difference of the direction angle distribution. A larger gradient direction offset indicates a greater difference in texture disorder assessment due to different enhancement paths. This significantly alters the perceived edge orientation of the region. The brightness enhancement magnitude offset is determined by calculating the standard deviation of the brightness values ​​at that point across multiple intermediate enhancement results. A large brightness enhancement magnitude offset indicates that the brightness enhancement / reduction magnitudes at that point are highly inconsistent across different enhancement paths. After normalizing the calculated offsets of the three features, a local enhancement stability index is generated. A higher local enhancement stability index value indicates greater consistency among the three features across different enhancement paths, suggesting more stable and reliable enhancement processing at that location. Conversely, a lower local enhancement stability index value indicates greater randomness or path dependence in the enhancement process, making the results unreliable.

[0037] Preferably, each pixel is divided into different confidence levels based on the numerical range of the local enhancement stability index. For example, a high confidence threshold and a low confidence threshold are set through statistical distribution or prior knowledge. A high confidence level is higher than the high confidence threshold, indicating that the enhancement is very stable; a medium confidence level is between the high confidence threshold and the low confidence threshold, indicating that the enhancement has some fluctuations; and a low confidence level is lower than the low confidence threshold, indicating that the enhancement is extremely unstable. Then, a matrix with the same size as the first enhanced image is generated. For each position in the image, the confidence level represented by each pixel is assigned to obtain a local enhancement confidence map, which intuitively marks the credible and uncredible enhancement areas in the first enhanced image.

[0038] Furthermore, step S500 also includes, after reading the current partitioning and block granularity, performing skin image repartitioning processing according to the correspondence between the partitioning and block granularity and the overall structural scale of the skin image, establishing macro partitions, wherein the area of ​​the macro partitions is larger than the area of ​​the local blocks; based on the macro partitions, performing regional continuity and response consistency verification on the enhancement results within each macro partition, establishing a macro response feature set characterizing enhancement stability at the macro scale; and performing joint mapping analysis between the macro response feature set and the local enhancement stability index to complete the credibility grading.

[0039] Preferably, the current partitioning granularity of the initial local blocks used to generate the first enhanced image is obtained, for example, with a side length of 64 pixels. Based on the correspondence between the partitioning granularity and the overall structural scale of the skin image, which may be a fixed multiple relationship or a ratio based on the image size, the scale of the macro partition is determined, and the skin image is repartitioned. That is, the scale of the macro partition is used to re-divide the skin image, including regular grid division of non-overlapping areas and sliding window division of overlapping areas, thereby determining the macro partitions. Each macro partition covers multiple original local blocks, that is, the area of ​​the macro partition is larger than the area of ​​the local blocks.

[0040] Preferably, for each macro-region, the enhancement results within it are verified for continuity and response consistency. Continuity verification within the region involves calculating the spatial autocorrelation or variation function of pixel intensity, gradient magnitude, or texture features within the macro-region to check for unnatural abrupt changes, patches, or artifacts in the enhanced image within the macro-region. If the enhancement result is natural and continuous, its spatial autocorrelation should be high and smoothly decay over short distances. If abnormal correlation breaks or severe local variance are detected, the continuity within the region is poor. For example, the entropy of the gradient magnitude histogram of pixel intensity within the region is calculated; an excessively high entropy value indicates the presence of numerous disordered edges, which may be noise or discontinuous artifacts. Alternatively, the variance of the average local contrast is calculated; an excessively large variance indicates uneven enhancement. Intra-regional response consistency verification refers to the classification and analysis of the enhancement results of all original local blocks covered by the macro-region. Specifically, for each macro-region, all original local blocks whose center points fall within that macro-region are identified. For each repeated sub-regional block, statistical features such as average brightness, average gradient, and texture entropy of its enhancement results are extracted. The distribution consistency of different statistical feature values ​​within the macro-region is analyzed, for example, by calculating the median absolute deviation or Gini coefficient of these feature values. A smaller median absolute deviation indicates higher intra-regional response consistency, suggesting statistical uniformity within the macro-region enhanced by different small blocks. Then, based on the results of intra-regional continuity verification and response consistency verification, a set of macro-response features characterizing the enhancement stability at the macro scale is established.

[0041] Preferably, a joint mapping analysis is performed between the macroscopic response feature set and the local enhancement stability index. Specifically, for each pixel, its macroscopic partition is identified and determined. If the continuity score within the macroscopic partition where the pixel is located is extremely low, it indicates the presence of severe artifacts or discontinuities. Even if the local enhancement stability index of the pixel is high, its final confidence level is downgraded. If the response consistency score and the continuity score within the macroscopic partition where the pixel is located are both high, and the local enhancement stability index of the pixel is also high, then the final confidence level of the pixel is improved or confirmed as the highest level. For points where the local enhancement stability index is at a critically medium level, their final classification is arbitrated by macroscopic features. For example, if the overall performance of the macroscopic region is good, these points are upgraded to medium-high confidence; if the performance of the macroscopic region is poor, they are downgraded to medium-low confidence. Finally, by combining the microscopic local enhancement stability index and the macroscopic response feature set, the final confidence level is calculated for each location to complete the confidence level classification.

[0042] Furthermore, step S500 also includes performing spatial continuity constraint processing on the credibility classification results of adjacent local regions when constructing the local enhanced credibility map, so as to suppress credibility abrupt changes caused by local noise or partition boundary effects.

[0043] Preferably, when generating a local enhancement confidence map, local noise interference and partition boundary effects can cause sudden changes in confidence. Local noise interference refers to the image acquisition noise or subtle artifacts introduced during the enhancement process, which may cause abnormal fluctuations in the stability index of a single pixel or a very small region, thereby producing isolated and unreasonable low confidence points or high confidence points. Partition boundary effects refer to the fact that when a damaged region feature happens to cross the partition boundary, since the enhancement functions of different partitions are independent of each other, discontinuous and jump-like confidence evaluation results may be generated on both sides of the boundary. Spatial continuity constraint processing is applied to the credibility grading results of adjacent local regions. This involves using neighborhood information to correct the credibility value of each region, making it more consistent with the surrounding environment. This ensures that the output credibility map conforms to physical and physiological common sense in its spatial distribution, and that the credibility changes of skin lesions and their enhancement results are gradual and continuous. Specifically, the local enhancement credibility map is treated as a grayscale image, with different credibility levels corresponding to different grayscale values. Mean filtering is used to traverse the image, replacing the credibility value of the center point of the window with the average value of all points within the window to quickly eliminate isolated noise points. Alternatively, Gaussian filtering is used, employing a Gaussian kernel for weighted averaging, with higher weights for neighboring points closer to the center. Median filtering is used to replace the center point value with the median of the credibility values ​​within the window to eliminate isolated outliers like salt-and-pepper noise while preserving edges. This ensures the physical rationality of the local enhancement credibility map is improved and suppresses artifacts introduced during the calculation process.

[0044] Furthermore, step S500 also includes: dividing the local regions in the first enhanced image into confidence states according to the local enhancement confidence map, and constructing high-confidence enhancement region identifiers, medium-confidence enhancement region identifiers, and low-confidence enhancement region identifiers; constructing constraints to maintain the continuity of the original skin texture structure for the low-confidence enhancement regions, and limiting the brightness stretching amplitude and gradient enhancement intensity, and performing directional enhancement processing of pressure damage-related response features; reading the enhancement results of the corresponding regions in the first enhanced image in the medium-confidence enhancement regions, and performing local fine-tuning processing based on the reading results; extracting enhancement parameter distribution features for the high-confidence enhancement regions, establishing enhancement reference constraints, and using the enhancement reference constraints to guide the secondary adaptive enhancement processing of the medium-confidence enhancement regions and low-confidence enhancement regions to construct a second enhanced image.

[0045] Preferably, the secondary adaptive enhancement process is triggered based on the local enhancement confidence map, that is, the secondary enhancement of the first enhanced image is performed with hierarchical optimization, focusing on repairing low confidence regions, fine-tuning medium confidence regions, and using high confidence regions as benchmarks to generate a second enhanced image with better overall quality. Specifically, according to the confidence level of each pixel position in the local enhancement confidence map, the local regions in the first enhanced image are divided into confidence states, that is, the pixels in the first enhanced image are clustered into three mutually exclusive connected region sets, including a high confidence region identifier mask, a medium confidence region identifier mask, and a low confidence region identifier mask.

[0046] Preferably, since low-confidence enhancement regions are unstable and may contain noise or artifacts, constraints are constructed to maintain the continuity of the original skin texture structure in these regions. This involves introducing a total variational regularization term during secondary enhancement, which penalizes the abnormally large gradient magnitude of the image, thereby maintaining edge smoothness and texture coherence. The brightness stretching amplitude is then limited, for example, the allowed brightness variation range does not exceed ±15% of the original region's average brightness. Simultaneously, the gradient enhancement intensity is limited, for example, setting an upper limit of 1.5 times for the gradient magnitude enhancement factor to avoid over-sharpening and introducing false edges. Then, targeted enhancement processing of pressure-related damage response features is performed. This involves limited enhancement of the features most relevant to pressure damage under the constructed constraints, such as a slight increase in the relative intensity of the local red channel or texture contrast in a specific direction, while suppressing changes in other irrelevant features to ensure that new artifacts are avoided to the greatest extent possible.

[0047] Preferably, the enhancement results of the medium confidence region are basically reliable. From the first enhanced image, all pixel values ​​covered by the medium confidence region identifier mask are extracted as the initial enhancement result of the region. Local fine-tuning is performed based on the reading results, which may include slight contrast optimization, smoothing for small patches, or brightness / color consistency adjustment with adjacent high confidence regions. The adjustment magnitude is set with a threshold to ensure that no drastic changes are made, such as applying mild adaptive histogram adjustment, the adjustment parameters of which are jointly determined by the local statistics of the region and the statistics of adjacent high confidence regions. High-confidence regions are standard regions where enhancement effects are stable and reliable. Analysis is performed on high-confidence regions and their corresponding original image regions, and inverse analysis is used to extract enhancement parameter distribution features, such as the local average shape of the brightness mapping curve, the distribution of gain coefficients for each color channel, the statistical value of local contrast stretching factors, and effective texture enhancement filters and their parameters. These extracted parameter features are then transformed into enhancement reference constraints for secondary enhancement of medium- and low-confidence regions. For example, the average brightness of enhanced medium- and low-confidence regions is required to approach the average brightness of the high-confidence region, and the local contrast statistical distribution of enhanced medium- and low-confidence regions is required to be as similar as possible to the distribution of the high-confidence region, using the color statistics of the high-confidence region as a reference target. Then, when performing adaptive enhancement processing on medium- and low-confidence regions, these enhancement reference constraints are incorporated, including using them as part of the optimization objective during directional enhancement of the low-confidence region and as the boundary for parameter selection when setting the parameter search range for fine-tuning processing of the medium-confidence region. This ensures that the generated second-enhanced image remains consistent, avoiding visual fragmentation between regions of different confidence levels. Finally, the enhancement processing results of the low, medium, and high-confidence regions are stitched together to form a complete second-enhanced image, improving the global consistency and security of image enhancement.

[0048] Step S600: Perform interactive fusion of the first enhanced image and the second enhanced image to output the third enhanced image.

[0049] Step S600 further includes, under the condition of spatial alignment, calculating the enhancement offset in the dimensions of local texture, gradient response, and brightness response based on the first enhanced image and the second enhanced image, and constructing a local enhancement offset field characterizing the second enhanced image relative to the first enhanced image; performing confidence modulation processing on the local enhancement offset field based on the local enhancement confidence map to form a confidence-constrained enhancement offset modulation field; using the first enhanced image as the enhancement master state structure, performing controlled injection processing on the enhancement offset modulation field to introduce pressure damage-related correction enhancement components in the second enhanced image; and reconstructing the third enhanced image based on the controlled injection processing result.

[0050] Preferably, the first and second enhanced images are interactively fused. Specifically, the first image serves as a stable base, and the improved portions of the second image, verified by credibility, are selectively absorbed. Specifically, the first and second enhanced images are spatially aligned to ensure a one-to-one correspondence between pixel positions. For each pixel position in the image, the changes in local texture, gradient response, and brightness response of the second enhanced image relative to the first enhanced image are calculated. Specifically, a small neighborhood is taken centered on each pixel position, and the texture features of the first and second enhanced images in that neighborhood are calculated. For example, the contrast of the gray-level co-occurrence matrix is ​​calculated, and the difference between the two texture features is calculated based on cosine distance or Euclidean distance as the texture enhancement offset at that point. The gradient magnitude of each pixel in the first and second enhanced images is calculated separately. The gradient difference between the second and first enhanced images is used as the gradient response enhancement offset, with a positive value indicating that the secondary enhancement makes the edge of the point sharper. The brightness value of each pixel in the first and second enhanced images is calculated, with the difference between the brightness values ​​of the second and first enhanced images used as the brightness response enhancement offset, with a positive value indicating that the secondary enhancement makes the point brighter. Then, the three enhancement offsets of each pixel are combined into a three-dimensional enhancement offset vector to determine the local enhancement offset field of the second enhanced image relative to the first enhanced image, and to fully and quantitatively describe the direction and intensity of the modification of the first enhanced image by each pixel position in each feature dimension.

[0051] Preferably, the local enhancement offset field is subjected to confidence modulation processing based on the local enhancement confidence map, including amplifying or releasing the enhancement offset corresponding to the low confidence enhancement region, i.e., multiplying it by a weight factor close to 1; suppressing or freezing the enhancement offset corresponding to the high confidence enhancement region, i.e., multiplying it by a weight factor close to 0; and the modulation processing of the enhancement offset corresponding to the medium confidence enhancement region is between the low confidence enhancement region and the high confidence enhancement region. The confidence value is mapped to a weight value in the interval [0, 1], and the local enhancement offset field is weighted point by point to obtain a confidence-constrained enhancement offset modulation field, which weakens the disturbance in the high confidence region and highlights the improvement suggestions in the low confidence region.

[0052] Preferably, the enhanced main structure refers to using the first enhanced image as the base and main structure for fusion, and performing controlled injection processing on the enhanced offset modulation field. This involves targeted image operations based on the type of offset. For brightness offset, a smooth brightness adjustment layer guided by the brightness response enhancement offset is added to the brightness channel of the first enhanced image. For gradient offset, the gradient response enhancement offset is treated as an edge sharpening intensity map to guide spatially adaptive sharpening of the first enhanced image, applying stronger sharpening to areas with high gradient response enhancement offset values. For texture offset, the local texture enhancement offset is converted into adjustments to local filter parameters, fine-tuning the texture representation of the first enhanced image to approximate the superior texture features in the second enhanced image. Next, pressure-related damage correction enhancement components in the second enhanced image are identified and introduced. For example, in a suspected erythema region, if the brightness offset enhancement is negative and has low confidence, it indicates that the first enhancement may have caused the erythema to become brighter and distorted. The second enhancement corrects this by making it darker, and the darkening correction component is injected with emphasis. After controlled adjustments to brightness, gradient, and texture, the modified image components are obtained, which are then recombined into a new image and subjected to color balancing, global contrast fine-tuning, etc., to ensure visual consistency. Finally, a clear and reliable third-enhanced image is output, thereby ensuring the accuracy of damage identification and the reliability of the enhancement results.

[0053] In the above text, refer to Figure 1 An image processing method for enhanced skin pressure injury monitoring according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 An image processing system for enhanced monitoring of skin pressure injuries according to an embodiment of the present invention is described.

[0054] The image processing system for enhanced skin pressure injury monitoring according to embodiments of the present invention addresses the technical problems in the prior art where conventional image enhancement cannot adapt to the local heterogeneity of skin images, is difficult to effectively extract early weak damage features, and has low reliability. It achieves the technical effect of accurately enhancing local micro-damage features to ensure the reliability of subsequent image annotation results. Figure 2 As shown, the image processing system for enhanced skin pressure injury monitoring includes: an image partitioning processing module 10, a block feature extraction module 20, an image enhancement processing module 30, a first enhanced image creation module 40, a second enhanced image creation module 50, and a third enhanced image output module 60.

[0055] Image partitioning processing module 10 is used to perform adaptive partitioning processing on the skin image to be processed, establishing N local blocks, wherein the N local blocks include repeated partitioning blocks; block feature extraction module 20 is used to extract the texture entropy, gradient distribution, and brightness deviation features of each local block, and construct a local micro-damage response map based on the extraction results; image enhancement processing module 30 is used to construct an adaptive enhancement function for each local block based on the local micro-damage response map, and apply different adaptive enhancement functions to the repeated partitioning blocks of the same original skin image for enhancement processing, constructing multiple sets of enhanced images; first The enhanced image creation module 40 is used by the interactive enhancement processing module to perform interactive verification of multiple sets of enhanced images, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, and perform interactive enhancement processing based on the calculation results to create a first enhanced image; the second enhanced image creation module 50 is used to perform offset residual analysis of the first enhanced image, construct a local enhancement confidence map, trigger secondary adaptive enhancement processing based on the local enhancement confidence map, and create a second enhanced image; the third enhanced image output module 60 is used to interactively fuse the first enhanced image and the second enhanced image to output a third enhanced image.

[0056] The specific configuration of the second enhanced image creation module 50 will be described in detail below. The second enhanced image creation module 50 further includes: calculating the response offset residuals of the same local region under different enhancement paths for the enhancement results generated corresponding to different repeated segmentation blocks in the first enhanced image, under the condition of spatial alignment; using the response offset residuals, calculating the local texture entropy offset, gradient direction offset, and brightness enhancement amplitude offset respectively, and combining the calculation results to generate a local enhancement stability index; using the local enhancement stability index to classify the credibility of the local region, forming the local enhancement credibility map.

[0057] The specific configuration of the second enhanced image creation module 50 will be described in detail below. The second enhanced image creation module 50 further includes: after reading the current segmentation and block granularity, performing skin image re-segmentation processing according to the correspondence between the segmentation and block granularity and the overall structural scale of the skin image to establish macro-partitions, wherein the area of ​​the macro-partitions is larger than the area of ​​the local blocks; based on the macro-partitions, performing regional continuity and response consistency verification on the enhancement results within each macro-partition to establish a macro-response feature set characterizing enhancement stability at the macro scale; and performing joint mapping analysis between the macro-response feature set and the local enhancement stability index to complete the credibility grading.

[0058] The specific configuration of the second enhanced image building module 50 will be described in detail below. The second enhanced image building module 50 further includes: dividing local regions in the first enhanced image into confidence states based on the local enhancement confidence map, and constructing high-confidence enhancement region identifiers, medium-confidence enhancement region identifiers, and low-confidence enhancement region identifiers; constructing constraints that maintain the continuity of the original skin texture structure for the low-confidence enhancement regions, and limiting the brightness stretching amplitude and gradient enhancement intensity, and performing directional enhancement processing of pressure damage-related response features; reading the enhancement results of the corresponding regions in the first enhanced image within the medium-confidence enhancement regions, and performing local fine-tuning processing based on the reading results; extracting enhancement parameter distribution features for the high-confidence enhancement regions, establishing enhancement reference constraints, and using the enhancement reference constraints to guide the secondary adaptive enhancement processing of the medium-confidence and low-confidence enhancement regions to construct the second enhanced image.

[0059] The specific configuration of the third enhanced image output module 60 will be described in detail below. The third enhanced image output module 60 further includes: under the condition of spatial alignment, calculating enhancement offsets in the dimensions of local texture, gradient response, and brightness response based on the first enhanced image and the second enhanced image, constructing a local enhancement offset field characterizing the second enhanced image relative to the first enhanced image; performing confidence modulation processing on the local enhancement offset field based on the local enhancement confidence map to form a confidence-constrained enhancement offset modulation field; using the first enhanced image as the enhancement master state structure, performing controlled injection processing on the enhancement offset modulation field to introduce pressure-related damage correction enhancement components from the second enhanced image; and reconstructing the third enhanced image based on the controlled injection processing result.

[0060] The specific configuration of the image partitioning module 10 will be described in detail below. The image partitioning module 10 further includes: the repeated partitioning blocks are generated by applying different partitioning strategies to the same skin image. The different partitioning strategies include different block sizes, different block starting offset positions, or different block shapes, so that the same image region is covered multiple times under different partitioning strategies, so as to construct repeated partitioning blocks with spatial overlap.

[0061] The specific configuration of the second enhanced image building module 50 will be described in detail below. The second enhanced image building module 50 further includes: when constructing a local enhanced confidence map, performing spatial continuity constraint processing on the confidence grading results of adjacent local regions to suppress confidence abrupt changes caused by local noise or partition boundary effects.

[0062] The image processing system for enhanced skin pressure injury monitoring provided in this embodiment of the invention can execute the image processing method for enhanced skin pressure injury monitoring provided in this embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An image processing method for enhanced monitoring of skin pressure injuries, characterized in that, The method includes: Adaptive partitioning is performed on the skin image to be processed to create N local blocks, wherein the N local blocks include repeated partitioning blocks; Extract the texture entropy, gradient distribution, and brightness deviation features of each local block, and construct a local micro-damage response map based on the extraction results; Based on the local micro-damage response map, an adaptive enhancement function is constructed for each local block. For repeated divisions of the same original skin image, different adaptive enhancement functions are applied to enhance the images, thus constructing multiple sets of enhanced images. Perform interactive verification of multiple sets of enhanced images, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, perform interactive enhancement processing based on the calculation results, and establish the first enhanced image; Perform offset residual analysis on the first enhanced image to construct a local enhancement confidence map. Based on the local enhancement confidence map, trigger a secondary adaptive enhancement process to establish a second enhanced image. The first and second enhanced images are interactively fused to output a third enhanced image.

2. The image processing method for enhanced skin pressure injury monitoring as described in claim 1, characterized in that, Perform offset residual analysis on the first enhanced image to construct a local enhancement confidence map, including: For the enhancement results generated corresponding to different repeated regions and blocks in the first enhanced image, under the condition of spatial alignment, the response offset residual of the same local region under different enhancement paths is calculated. Using the response offset residual, the local texture entropy offset, gradient direction offset, and brightness enhancement amplitude offset are calculated respectively, and the calculation results are combined to generate a local enhancement stability index. The credibility level of a local region is classified using the local enhancement stability index to form the local enhancement credibility map.

3. The image processing method for enhanced skin pressure injury monitoring as described in claim 2, characterized in that, The method of using the local enhancement stability index to classify the credibility of local regions also includes: After reading the current partitioning and block granularity, the skin image is repartitioned according to the correspondence between the partitioning and block granularity and the overall structural scale of the skin image, and macro partitions are established, wherein the area of ​​the macro partition is larger than the area of ​​the local block. Based on the macro-partitions, the continuity and response consistency of the enhancement results within each macro-partition are verified, and a set of macro-response features characterizing the stability of enhancements at the macro scale is established. The set of macroscopic response features is jointly mapped and analyzed with the local enhanced stability index to complete the credibility classification.

4. The image processing method for enhanced skin pressure injury monitoring as described in claim 1, characterized in that, Based on the local enhancement confidence map, a secondary adaptive enhancement process is triggered to establish a second enhanced image, including: Based on the local enhancement confidence map, the local regions in the first enhanced image are divided into confidence states, and high-confidence enhancement region identifiers, medium-confidence enhancement region identifiers, and low-confidence enhancement region identifiers are constructed. For low-confidence enhancement regions, constraints are constructed to maintain the continuity of the original skin texture structure, and the brightness stretching amplitude and gradient enhancement intensity are limited to perform targeted enhancement processing of pressure damage-related response features; Read the enhancement result of the corresponding region in the first enhanced image of the credible enhanced region, and perform local fine-tuning based on the reading result; The distribution features of enhancement parameters are extracted from the high-confidence enhancement region, and enhancement reference constraints are established. The enhancement reference constraints are used to guide the secondary adaptive enhancement processing of the medium-confidence enhancement region and the low-confidence enhancement region to construct the second enhanced image.

5. The image processing method for enhanced skin pressure injury monitoring as described in claim 1, characterized in that, The first and second enhanced images are interactively fused to output a third enhanced image, including: Under the condition of spatial alignment, the enhancement offset in the dimensions of local texture, gradient response and brightness response is calculated based on the first enhancement image and the second enhancement image, and a local enhancement offset field representing the second enhancement image relative to the first enhancement image is constructed. The local enhancement offset field is subjected to confidence modulation processing based on the local enhancement confidence map to form an enhancement offset modulation field constrained by confidence. Using the first enhanced image as the enhanced master state structure, a controlled injection process is performed on the enhanced offset modulation field to introduce a pressure damage-related correction enhancement component from the second enhanced image; Based on the results of controlled injection processing, a third enhanced image is reconstructed.

6. The image processing method for enhanced skin pressure injury monitoring as described in claim 1, characterized in that, The repeated segmentation blocks are generated by applying different segmentation strategies to the same skin image. The different segmentation strategies include different block sizes, different block starting offset positions, or different block shapes, so that the same image region is covered multiple times under different segmentation strategies, so as to construct repeated segmentation blocks with spatial overlap.

7. The image processing method for enhanced skin pressure injury monitoring as described in claim 1, characterized in that, When constructing the local enhanced credibility map, spatial continuity constraints are applied to the credibility classification results of adjacent local regions to suppress credibility abrupt changes caused by local noise or partition boundary effects.

8. An image processing system for enhanced monitoring of skin pressure injuries, characterized in that, The system is used to implement the image processing method for enhanced skin pressure injury monitoring according to any one of claims 1 to 7, the system comprising: The image partitioning module is used to perform adaptive partitioning processing on the skin image to be processed, and to establish N local blocks, wherein the N local blocks include repeated partitioning blocks; The block feature extraction module is used to extract the texture entropy, gradient distribution and brightness deviation features of each local block, and construct a local micro-damage response map based on the extraction results; The image enhancement processing module is used to construct an adaptive enhancement function for each local block based on the local micro-damage response map, and to apply different adaptive enhancement functions to the repeated divisions of the same original skin image to perform enhancement processing, thereby constructing multiple sets of enhanced images. The first enhanced image establishment module is used by the interactive enhancement processing module to perform interactive verification of multiple sets of enhanced images, calculate the consistency score of the local micro-damage response corresponding to each repeated enhanced image, perform interactive enhancement processing based on the calculation results, and establish the first enhanced image. The second enhanced image building module is used to perform offset residual analysis of the first enhanced image, construct a local enhancement confidence map, trigger secondary adaptive enhancement processing based on the local enhancement confidence map, and build the second enhanced image; The third enhanced image output module is used to interactively fuse the first enhanced image and the second enhanced image to output the third enhanced image.

Images