Skin rejuvenation treatment effect texture image analysis method and system

Abstract

The application discloses a skin rejuvenation treatment effect texture image analysis method and system, and relates to the technical field of medical image processing, and comprises the following steps: acquiring treatment texture images and illumination collection data of a target object; performing separation operation on the treatment texture images according to a treatment sequence, and performing symmetry consistency evaluation on the illumination collection data and the separation operation result to generate a stable mask graph; respectively performing weighted alignment on the separation operation result before and after treatment according to the stable mask graph, generating a locked texture graph in pairs, and performing symmetry statistics on the locked texture graph to generate a brightness reference graph. According to the scheme, the weighted alignment based on the stable mask graph and the symmetry statistics of the brightness reference graph are used, so that the texture comparison before and after treatment is changed from 'pixel comparison' to 'brightness comparison', the interference of illumination drift on difference calculation is significantly reduced, a normalized texture graph is generated by fusing and correcting illumination collection parameters, and an alignment difference bottom graph is obtained by modulating the stable mask graph.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method and system for analyzing texture images of skin rejuvenation treatment effects. Background Technology

[0002] With the development of optoelectronic medicine and medical image processing technology, energy-based skin rejuvenation treatments have been widely used in the field of skin improvement. These treatments typically use energy forms such as lasers, intense pulsed light, radiofrequency, or ultrasound to act on the surface and superficial structures of the skin in order to improve the skin texture. Correspondingly, the evaluation of treatment effects has gradually evolved from simple visual observation to relying on skin surface texture images obtained by dermoscopy, high-magnification macro imaging equipment, etc., and combining image processing methods to analyze changes in skin texture.

[0003] In existing technologies, image analysis schemes for skin rejuvenation treatment effects typically involve acquiring skin surface texture images before and after treatment, then performing illumination correction, contrast enhancement, or simple registration processing on these images. Texture features are then extracted, and the treatment effect is determined through before-and-after difference analysis or feature comparison. These schemes can reflect macroscopic changes in the skin surface to a certain extent and have advantages such as relatively direct implementation paths and low computational complexity. However, the registration in these schemes often focuses on geometric alignment, lacking explicit constraints on stable and registrable regions. This leads to easy drift or mismatch of matching points in areas with weak texture, high noise, or reflectivity, resulting in unreasonable distortions during deformation alignment. Furthermore, such mismatches are directly passed to subsequent difference analysis, further making it difficult to distinguish between actual treatment-related changes and structural differences caused by registration errors. Consequently, the improved bright areas appear as stripes or blocks with a weak correspondence to wrinkles, grooves, and pore structures, leading to insufficient accuracy in the generated analysis report. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for analyzing the texture images of skin rejuvenation treatment effects, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: In a first aspect, this invention discloses a method for analyzing the texture image of skin rejuvenation treatment effects, comprising the following steps: Acquire the treatment texture image and lighting acquisition data of the target object; Separation operations are performed on the treatment texture image according to the treatment sequence, and the symmetry consistency of the illumination acquisition data and separation operation results is evaluated to generate a stable mask image; The separation operation results are weighted and aligned before and after treatment based on the stable mask image to generate locked texture images in pairs. Symmetrical statistics are then performed on the locked texture images to generate a brightness reference image. The brightness reference map and the locked texture map are fused and corrected according to the illumination acquisition parameters to generate a normalized texture map. The normalized texture map is then modulated with the stable mask map after gradient difference calculation to generate an aligned difference base map. The normalized texture map is subjected to multi-order differential weighting based on the illumination acquisition parameters to generate a texture difference map, and the texture difference map is combined with the alignment difference base map to perform fractional mapping to generate a response consistency map. The response consistency graph is subjected to consistent gating encoding to generate a structure index graph; Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map. The improved saliency map and the normalized texture map are then image-stitched together to generate an analysis report.

[0006] Secondly, this invention discloses a texture image analysis system for skin rejuvenation treatment effects, comprising: The data acquisition module is used to acquire the treatment texture image and lighting data of the target object; The stability analysis module is used to perform separation operations on the treatment texture image according to the treatment sequence, and to perform symmetry consistency evaluation on the illumination acquisition data and separation operation results to generate a stable mask image; The brightness analysis module is used to perform weighted alignment of the separation operation results before and after treatment based on the stable mask image, generate locked texture images in pairs, and perform symmetric statistics on the locked texture images to generate a brightness reference image. The difference analysis module is used to perform fusion correction on the brightness reference map and the locked texture map according to the illumination acquisition parameters to generate a normalized texture map, and modulate the normalized texture map with the stable mask map after gradient difference calculation to generate an aligned difference base map. The response analysis module is used to perform multi-order differential weighting on the normalized texture map according to the illumination acquisition parameters to generate a texture difference map, and to perform fractional mapping on the texture difference map in combination with the alignment difference base map to generate a response consistency map. The analysis report generation module is used to perform consistent gating encoding on the response consistency graph to generate a structure index graph; Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map. The improved saliency map and the normalized texture map are then image-stitched together to generate an analysis report.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This solution transforms the texture comparison before and after treatment from "pixel-comparable" to "brightness-comparable" by using weighted alignment based on a stable mask map and symmetrical statistics of the brightness baseline map. This significantly reduces the interference of illumination drift on the difference calculation. The solution generates a normalized texture map by fusing and correcting illumination acquisition parameters and modulates it with a stable mask map to obtain an aligned difference base map. This can centrally expose and isolate geometric mismatches and illumination residuals. The response consistency map and structural index map obtained by multi-order difference weighting and fractional mapping can limit the significance of improvement to the stable structural region, thereby improving the detectability, interpretability, and verification consistency of fine wrinkles and pore changes, and outputting a directly comparable image analysis report.

[0008] 2. This scheme solves the horizontal and vertical gradients of the normalized texture map separately and fuses it pixel by pixel with the brightness baseline map under a stable mask map. This can unify the brightness baseline in the stable region, highlight the edges of the real texture, and suppress the pseudo gradients introduced by reflections and shadows. The matching confidence map and the locked boundary map are fused pixel by pixel to generate a deformation constraint map. This tightens the deformation degrees of freedom at the boundary transition and relaxes the deformation intensity at high confidence, avoiding unreasonable distortion. Based on the deformation constraint map, the normalized texture map after treatment is resampled to generate a registration texture map, achieving accurate alignment in the same area and reducing structural drift. Attached Figure Description

[0009] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 A flowchart illustrating the steps of the texture image analysis method for skin rejuvenation treatment effects provided by this invention; Figure 2 This is a schematic diagram of the process for generating a stable mask image provided by the present invention; Figure 3 This is a schematic diagram of the process for generating a normalized texture map provided by the present invention; Figure 4 This is a schematic diagram of the process for generating an alignment difference base map provided by the present invention; Figure 5 This is a schematic diagram of the module functions of the texture image analysis system for skin rejuvenation treatment effects provided by the present invention. Detailed Implementation

[0010] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0011] Application Overview: In the image analysis of the effects of skin rejuvenation treatments, existing technologies lack explicit constraint mechanisms for stable and registable regions during the registration of skin texture images before and after treatment. This leads to drift or mismatch in feature point matching in low-texture areas, high-noise environments, or under reflective conditions, resulting in non-physical distortion during the deformation alignment stage. This registration error is propagated to the subsequent differential analysis process, making it difficult to distinguish between real treatment-related changes and registration artifacts. Ultimately, the generated improved highlight areas exhibit an unstructured distribution, weakening the correspondence with skin wrinkles, grooves, and pore structures, thus affecting the reliability of the analysis report.

[0012] For example, in the scenario of acquiring skin texture images in the facial nasal and forehead areas, the sebaceous gland secretion causes significant local reflection, making it difficult for existing registration algorithms to stably identify feature points in this area. This results in a large number of mismatched point pairs, causing abnormal stretching of the deformation field at the nasal grooves. The improvement area that should be smooth in the difference image appears as discrete striped artifacts that do not match the actual wrinkle direction, thus causing deviation in the evaluation of treatment effect based on the difference results.

[0013] If the above problems are not resolved, mismatches during the registration process will persist, resulting in a large number of spurious change signals caused by registration errors being mixed into the differential analysis results. This makes it difficult for the treatment effect assessment to accurately reflect the actual improvement in skin structure, thereby reducing the credibility of clinical decisions and potentially misleading the adjustment of subsequent treatment plans.

[0014] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0015] Example 1: Please see Figure 1 A texture image analysis method for skin rejuvenation treatment effects includes the following steps: Acquire the treatment texture image and lighting acquisition data of the target object; Separation operations are performed on the treatment texture image in the order of treatment, and the symmetry consistency of the illumination acquisition data and separation operation results is evaluated to generate a stable mask image; Based on the stable mask image, the separation operation results are weighted and aligned before and after treatment, and locked texture images are generated in pairs. Symmetric statistics are then performed on the locked texture images to generate a brightness reference image. Based on the illumination acquisition parameters, the brightness reference map and the locked texture map are fused and corrected to generate a normalized texture map. The normalized texture map is then modulated with the stable mask map after gradient difference calculation to generate an aligned difference base map. Based on the illumination acquisition parameters, the normalized texture map is subjected to multi-level differential weighting to generate a texture difference map. Then, the texture difference map is combined with the alignment difference base map to perform fractional mapping to generate a response consistency map. Perform consistent gating encoding on the response consistency graph to generate a structure index graph; Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map. The improved saliency map and the normalized texture map are then image-stitched together to generate an analysis report.

[0016] Among them, therapeutic texture image refers to the set of skin surface texture pixel matrix data directly acquired and output by an image acquisition device at at least two different treatment time points for the same target object and the same skin area during the skin rejuvenation treatment process; Illumination acquisition data refers to the set of raw acquisition parameter data directly output by the image acquisition device at the same acquisition moment when acquiring the therapeutic texture image, which is used to characterize the illumination state of the image at that time. The separation operation result refers to the set of paired image data obtained after performing temporal separation processing on the therapeutic texture image according to the image reception order; A stable mask image is a two-dimensional weighted image with the same size as the treatment texture image, calculated based on the pre-treatment and post-treatment images of the treatment texture image and combined with illumination acquisition data, in the same pixel coordinate system. Locked texture maps refer to paired texture image data obtained by performing weighted decision and intensity alignment processing on pre-treatment and post-treatment texture images respectively, using stable mask images as pixel-level constraints, and used for subsequent illumination uniformity and difference analysis. A brightness reference image is a pixel-level brightness reference image obtained by symmetrical statistical calculation based on the pixel grayscale distribution of two locked texture images before and after treatment, under the constraint of a stable mask. Normalized texture map refers to a set of paired texture images that are formed after performing brightness scale unification, contrast uniformization and grayscale distribution alignment on the locked texture map under the overall constraint of illumination acquisition parameters. It is used to reflect the skin texture before and after treatment under a unified illumination and brightness benchmark. Alignment difference base map refers to an image-type difference distribution data generated based on the pre-treatment normalized texture map and the post-treatment normalized texture map, under the premise of spatial alignment, by calculating pixel-level gradient differences and combining them with stable mask modulation. A texture difference map is a pixel-level difference representation image generated by performing multi-order difference operations and weighted fusion on a normalized texture image before and after treatment, based on the degree of texture change between the normalized texture image before treatment and the normalized texture image after treatment at the same spatial location. A response consistency map is a pixel-level consistent distribution image data generated by performing a consistency mapping operation on an alignment difference base map after performing multi-order difference weighting on a normalized texture map to obtain the texture difference result. A structure indexed graph is a pixel-level structure constraint indexed image generated based on a response consistency graph. An improvement saliency map refers to two-dimensional image data formed by spatially focusing the intensity of changes in skin texture before and after treatment under structural constraints, which is used to characterize the degree of significance of texture changes. An analysis report refers to a multi-layered composite image data set formed by image stitching and overlay under the same spatial coordinate system.

[0017] This scheme generates a stable mask image by separating and evaluating the symmetry consistency of the treatment texture image. It uses illumination constraints to suppress false differences caused by reflected shadows, locks high consistency regions to reduce misjudgment, generates a locked texture image by weighted alignment based on the stable mask image, and generates a brightness reference image by symmetric statistics. This achieves alignment of the front and back regions at the same scale and establishes a unified brightness reference, reducing the interference of brightness drift on texture quantization. It also generates a normalized texture image by fusion correction and generates an aligned difference base map by gradient difference modulation. Under a unified brightness baseline, structural differences are highlighted and suspicious mismatch areas are identified to provide a basis for subsequent gating. Multi-order differential weighting generates a texture difference map and fractional mapping generates a response consistency map, which enhances the sensitivity to micro-texture changes and compresses the influence of extreme noise, resulting in a consistent distribution of changes before and after. Consistency gating encoding generates a structural index map: the consistency is spatially encoded into an index expression of a constrained structure, which facilitates stable positioning. The structural index map constrains the alignment of the difference base map difference enhancement to generate an improvement saliency map, which is then stitched together to generate an analysis report. This allows the saliency to focus on the vicinity of credible structures and presents the improvement and inconsistency positions intuitively with a comparison map, improving interpretability and verifiability.

[0018] The above describes a complete scheme for analyzing the texture image of skin rejuvenation treatment effects. The following section details the acquisition of treatment texture images and illumination data for the target object, specifically including: The treatment texture image of the target object is obtained by dermatoscopy; the treatment texture image includes, but is not limited to, magnified skin texture image, pre-treatment texture image and post-treatment texture image, etc. Illumination data of the target object is acquired through a light acquisition device; the light acquisition data includes, but is not limited to, light intensity related parameter vectors, exposure related parameter vectors, gain related parameter vectors, white balance related parameter vectors, and overall acquisition condition vectors, etc. Perform separation operations on the treatment texture image in the order of treatment: The treatment texture image is input into the image buffer as an image sequence arranged chronologically. The pixel matrix of the first image is read in sequence index and output as the pre-treatment texture image. At the same time, the pixel matrix of the second image is read in sequence index and output as the post-treatment texture image. The pre-treatment texture image and the post-treatment texture image are combined to obtain the separation operation result.

[0019] The above describes how to acquire the treatment texture image and lighting data of the target object. The following describes how to perform a symmetric consistency evaluation on the lighting data and separation operation results to generate a stable mask image. Please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram of the process for generating a stable mask image provided in an embodiment of this application. Generating a stable mask image specifically includes: The separation operation results were analyzed by pixel-wise grayscale difference before and after treatment. The illumination acquisition parameters are solved by parameter amplitude calculation, and the gray-scale difference analysis results are normalized by mean and fractional mapping based on the parameter amplitude calculation results to generate a stable mask image.

[0020] Among them, the parameter amplitude solution refers to the single scalar parameter value obtained after performing a unified numerical mapping operation on the overall vector of illumination acquisition parameters, which is used to characterize the influence intensity of the overall illumination conditions on the image grayscale stability during the current image acquisition. Gray-level difference analysis results refer to the difference matrix formed by comparing the gray-level intensity of the separated pre-treatment texture image and post-treatment texture image at the same pixel coordinate position.

[0021] The above content will be described in detail below: The separation operation results were analyzed by pixel-wise grayscale difference before and after treatment: The "pre-treatment texture map" and "post-treatment texture map" are converted into grayscale matrix representations of the same size. The absolute difference is calculated for the same pixel position to obtain the pixel difference value. Then, the pixel difference values ​​of all pixel positions are arranged according to the original coordinates to form a grayscale difference matrix. Amplitude normalization is performed on the gray-level difference matrix to generate gray-level difference analysis results; The illumination acquisition parameters are solved for amplitude, and the grayscale difference analysis results are normalized by mean and fractional mapping based on the amplitude solution to generate a stable mask image. The illumination acquisition parameters are solved to generate parameter amplitude scalars. The calculation process of parameter amplitude scalars is as follows: the illumination acquisition parameters are regarded as parameter vectors composed of multiple components. The square operation is performed on each component, and the square results are summed to obtain the sum of squares. Then, the square root operation is performed on the sum of squares to obtain the parameter amplitude scalars. The parameter amplitude scalars are used as the parameter amplitude solution results. Based on the parameter amplitude scalar, the gray-level difference analysis results are subjected to parameter modulation and fractional mapping to generate a stable mask image. The calculation process of parameter modulation and fractional mapping is as follows: First, the modulation difference value is calculated for each pixel position. The modulation difference value = gray-level difference analysis result × (parameter amplitude scalar / (parameter amplitude scalar + constant -)). Then, fractional mapping is performed on the modulation difference value to obtain the mask value. The mask value = constant - / (constant - + modulation difference value). The mask values ​​of all pixel positions are combined to form a mask value matrix as the output of the stable mask image.

[0022] This scheme performs grayscale difference analysis on the separation operation results before and after treatment, pixel by pixel. It explicitly quantifies the brightness change of texture at the same pixel coordinates before and after treatment, forming a distinguishable difference characterization of local reflections, shadows and real texture changes, reducing misjudgments caused by overall brightness fluctuations alone. It performs parameter amplitude calculation on the illumination acquisition parameters, and performs mean normalization and fractional mapping on the grayscale difference analysis results based on the parameter amplitude calculation results to generate a stable mask map. It scales the difference amplitude according to the acquisition condition intensity and compresses the difference into stability weights, so that the position with stronger illumination disturbance has a lower weight and the stable region has a higher weight, thus providing robust and effective region constraints for subsequent processing.

[0023] The above describes the symmetric consistency evaluation of illumination acquisition data and separation calculation results to generate a stable mask image. The following describes the fusion correction of the brightness reference image and locked texture image based on illumination acquisition parameters to generate a normalized texture image. Please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram of the process for generating a normalized texture map provided in an embodiment of this application. Generating a normalized texture map specifically includes: Within the stable mask image, the parameter amplitude solution results corresponding to the illumination acquisition parameters are gated and fused with the brightness reference image to generate a correction coefficient image. The correction coefficient map is multiplied pixel by pixel with the locked texture map to generate a coefficient modulation map. Then, pixel-level weighted fusion is performed on the coefficient modulation map and the brightness reference map to generate a normalized texture map.

[0024] Among them, the correction coefficient map refers to the pixel-level coefficient distribution image that corresponds one-to-one with the locked texture map in terms of spatial size, which is calculated based on the illumination acquisition parameters and the brightness reference map. A coefficient modulation map refers to intermediate image data obtained by performing pixel-by-pixel multiplication operations using a correction coefficient map and a locked texture map as the only input data.

[0025] The above content will be described in detail below: Based on the stable mask image, the separation operation results are weighted and aligned before and after treatment to generate a locked texture image. Then, symmetric statistics are performed on the locked texture image to generate a brightness baseline image. For each pixel location, the grayscale value of the pre-treatment texture image is multiplied by the mask value at the same position in the stable mask image to obtain the pre-treatment weighted image. The calculation process is "pre-treatment weighted image = pre-treatment texture image × stable mask image". Similarly, the grayscale value of the post-treatment texture image is multiplied by the mask value at the same position in the stable mask image to obtain the post-treatment weighted image. The calculation process is "post-treatment weighted image = post-treatment texture image × stable mask image". Within the area covered by the stable mask image, calculate the weighted mean and weighted variance of the pre-treatment weighted image and the post-treatment weighted image, respectively. The weighted mean is calculated as follows: "Weighted mean = Σ(pixel gray level × mask value) / [Σ(mask value) + constant - ]", and the weighted variance is calculated as follows: "Weighted variance = Σ(mask value × (pixel gray level - weighted mean)". 2 The formula is: ) / [Σ(mask value)+constant-]”, and the weighted image before treatment and the weighted image after treatment are linearly normalized with weighted mean and weighted variance respectively to complete the intensity scale alignment. The calculation process of linear normalization is: "aligned image = (weighted image - weighted mean) / √(weighted variance + constant-)”, so as to obtain the locking image before treatment and the locking image after treatment respectively, and output the two together as the locking texture image; Within the area covered by the stable mask image, the mask-weighted average brightness of the pre-treatment locked image and the post-treatment locked image are calculated separately. The calculation process is "average brightness = Σ(locked image pixel × mask value) / [Σ(mask value) + constant 1]". Then, the two are symmetrically combined to obtain the common average brightness. The calculation process is "common average brightness = (previous average brightness + postvious average brightness) / constant 2". Subsequently, a reference brightness matrix is ​​constructed at the pixel level. The calculation process is "pixel symmetric average = (previous locked image + post-treatment locked image) / constant 2". The pixel symmetric average is multiplied pixel by pixel with the stable mask image to obtain the mask reference brightness matrix. The calculation process is "mask reference brightness = pixel symmetric average × stable mask image". Finally, the mask reference brightness matrix and the common average brightness are linearly mapped to output the brightness reference image. The calculation process of the linear mapping is "brightness reference image = mask reference brightness × constant 1 + common average brightness × (constant 1 - stable mask image)". Within the stable mask image, the parameter amplitude solutions corresponding to the illumination acquisition parameters are gated and fused with the brightness reference image to generate a correction coefficient map: The parameter gating coefficient is constructed using the parameter amplitude scalar. The calculation process of the parameter gating coefficient is "parameter gating coefficient = constant - / (constant - + parameter amplitude)" and the parameter gating coefficient is expanded into a parameter gating matrix of the same size as the brightness reference map. The luminance reference map is gated in the domain by the stable mask map to obtain the mask reference matrix. The calculation process of the mask reference matrix is ​​"mask reference matrix = luminance reference map × stable mask map". The mask mean reference is calculated for the mask reference matrix within the coverage area of ​​the stable mask map. The calculation process of the mask mean reference is "mask mean reference = Σ(mask reference matrix) / [Σ(stable mask map) + constant -]". Calculate the pixel-level reference ratio matrix. The calculation process of the reference ratio matrix is ​​"Reference ratio matrix = Mask mean reference / (Brightness reference map + constant -)". Then, perform gated fusion of the reference ratio matrix and the parameter gate matrix to obtain the fusion coefficient matrix. The calculation process of the fusion coefficient matrix is ​​"Fusion coefficient matrix = Reference ratio matrix × Parameter gate matrix". The fusion coefficient matrix and the stable mask map are subjected to intra-domain constraints to form the final coefficient distribution. The calculation process of the final coefficient distribution is "correction coefficient map = fusion coefficient matrix × stable mask map". The correction coefficient map is then linearly normalized according to the minimum and maximum values ​​to obtain the correction coefficient map. The correction coefficient map is multiplied pixel-by-pixel with the locked texture map to generate a coefficient modulation map. Then, a pixel-level weighted fusion is performed between the coefficient modulation map and the brightness reference map to generate a normalized texture map. The specific calculation formula is as follows: ; In the formula, Represents the normalized texture map in pixels Output pixel value at that location, Represents pixel-level fusion weights. Represents the coefficient modulation map in pixels Pixel value at that location, Indicates the brightness reference map at the pixel The pixel values ​​at each location are normalized during the calculation process.

[0026] This scheme, within a stable mask image, performs gated fusion of the parameter amplitude solution corresponding to the illumination acquisition parameters and the brightness reference image to generate a correction coefficient image. This ensures that the correction amount is adaptively allocated only within areas with consistent illumination and reliable texture, and is synchronously constrained by the acquisition conditions, reducing brightness drift introduced by reflections and shadows. The correction coefficient image is multiplied pixel-by-pixel with the locked texture image to generate a coefficient modulation image, which is then fused with the brightness reference image at the pixel level to generate a normalized texture image. This unifies the brightness and contrast baselines before and after treatment and suppresses noise amplification, improving the comparability of subtle texture differences in the same area and the stability of subsequent registration and matching.

[0027] The above describes the fusion correction of the brightness reference map and the locked texture map based on the illumination acquisition parameters to generate a normalized texture map. The following describes the modulation of the normalized texture map with a stable mask map after gradient difference calculation to generate an aligned difference base map. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the process for generating an alignment difference basemap provided in an embodiment of this application. Generating the alignment difference basemap specifically includes: Gradients are calculated for the normalized texture map in both the horizontal and vertical directions, and the gradient results are then fused pixel-by-pixel with the brightness reference map under a stable mask. The pixel-by-pixel fusion results are subtracted and modulated by a normalized consistency map to obtain a matching confidence map. The matching confidence map and the locked boundary map are then fused pixel by pixel to generate a deformation constraint map. Among them, the normalized consistent map is obtained by performing differential mapping on the normalized texture map; The locked boundary map is obtained by performing neighborhood difference on the stable mask map; Based on the deformation constraint map, the normalized texture map is resampled after treatment to generate a registration texture map. The grayscale difference between the registration texture map and the stable mask map is analyzed to generate an alignment difference base map.

[0028] The gradient solution result refers to the set of directional gradient response data obtained by performing local gray-level change rate calculation on the normalized texture map along the horizontal and vertical directions, using the normalized texture map as the only input image data. The pixel-by-pixel fusion result refers to the pixel-level structural response matrix formed by numerically coupling the directional gradient information calculated from the normalized texture map with the brightness reference map under the constraint of the stable mask map at the same pixel coordinate position. A normalized consistent image is a pixel-level consistent distribution image generated based on a normalized texture map, used to characterize the similarity of normalized textures in the same spatial location before and after treatment. A matching confidence map is a two-dimensional image data that is calculated based on a normalized texture map, a brightness reference map, a stable mask map, and a normalized consistency map, and is used to characterize the reliability of the spatial matching of textures before and after treatment. Locked boundary map is a type of image-based intermediate data calculated from a stable mask map through spatial neighborhood difference, used to describe the spatial transition intensity distribution between stable and unstable regions in the stable mask map; Deformation constraint map refers to pixel-level spatial constraint image data used to limit the deformation intensity and deformation degrees of freedom that can occur at each pixel position when performing geometric resampling on a normalized texture map; A registration texture map refers to paired image data formed by performing spatial resampling operations on the post-treatment texture image in the normalized texture map under the spatial constraints of the deformation constraint map, and aligning it with the pre-treatment texture image in the normalized texture map in the same pixel coordinate system. This data is used to characterize the alignment relationship of the texture structure in the same region before and after treatment.

[0029] The above content will be described in detail below: The gradient of the normalized texture map is calculated separately in the horizontal and vertical directions, and the gradient results are then fused pixel-by-pixel with the brightness reference map under a stable mask map. For each normalized image in the normalized texture map, calculate the horizontal gradient matrix and the vertical gradient matrix. The horizontal gradient matrix is ​​calculated by taking the absolute value of the gray level difference between two adjacent pixels in the same row for each pixel position, i.e., "Horizontal gradient = |Current pixel gray level - Right adjacent pixel gray level|". The vertical gradient matrix is ​​calculated by taking the absolute value of the gray level difference between two adjacent pixels in the same column for each pixel position, i.e., "Vertical gradient = |Current pixel gray level - Lower adjacent pixel gray level|". Then, take the square root of the sum of the squares of the horizontal and vertical gradient matrices at the same pixel position to obtain the gradient magnitude matrix, which is calculated as "Gradient magnitude = √(Horizontal gradient) / ... 2 +Vertical gradient 2 The gradient magnitude matrix is ​​then output as the gradient solution result. Under the constraint of a stable mask image, the gradient solution results are fused with the brightness reference image pixel by pixel to generate a pixel-by-pixel fusion result. The calculation process of pixel-by-pixel fusion is as follows: first, the mask complement matrix "complementary mask = 1 - stable mask image" is calculated, then the fused pixel value "fusion result = gradient magnitude × stable mask image + brightness reference image × complementary mask" is calculated, and the fusion result is linearly normalized according to the minimum and maximum values ​​and output as a pixel-by-pixel fusion result. The pixel-by-pixel fusion results are subtracted and modulated by a normalized consistency map to obtain a matching confidence map. The matching confidence map and the locked boundary map are then fused pixel by pixel to generate a deformation constraint map. Among them, the normalized consistent map is obtained by performing differential mapping on the normalized texture map; The locked boundary map is obtained by performing neighborhood difference on the stable mask map: Pixel-level difference mapping is performed on the pre-treatment and post-treatment normalized images in the normalized texture map to obtain the difference matrix. The difference matrix is ​​calculated as "Difference matrix = |Pre-treatment normalized image - Post-treatment normalized image|". Linear normalization is then performed on the difference matrix to obtain the normalized difference matrix. The linear normalization is calculated as "Normalized difference matrix = (Difference matrix - Minimum difference) / (Maximum difference - Minimum difference + Constant)". Finally, fractional mapping is performed on the normalized difference matrix to obtain the consistency matrix and output as a normalized consistent map. The fractional mapping is calculated as "Normalized consistent map = Constant / (Constant + Normalized difference matrix)". Based on the pixel-by-pixel fusion results, a matching confidence map is generated by subtracting the difference and modulating with a normalized consistency map. A pixel-level absolute difference is performed between the pre-treatment and post-treatment fusion response images to obtain a fusion difference matrix, calculated as "fusion difference matrix = |pre-treatment fusion response image - post-treatment fusion response image|". The fusion difference matrix is ​​then multiplied pixel-level with the normalized consistency map to obtain a consistency modulation difference matrix, calculated as "consistency modulation difference matrix = fusion difference matrix × normalized consistency map". Linear normalization is then performed on the consistency modulation difference matrix to obtain a confidence difference matrix, calculated as "confidence difference matrix = (consistency modulation difference matrix - minimum value) / (maximum value - minimum value + constant -)". Finally, a fractional mapping is performed on the confidence difference matrix to output a confidence distribution matrix as the matching confidence map, calculated as "matching confidence map = constant - / (constant - + confidence difference matrix)". Neighborhood difference is performed on the stable mask image to obtain the boundary magnitude matrix. The calculation process of neighborhood difference is as follows: "For each pixel position, calculate the absolute value of the difference between it and the left neighbor pixel, the absolute value of the difference between it and the right neighbor pixel, the absolute value of the difference between it and the upper neighbor pixel, and the absolute value of the difference between it and the lower neighbor pixel, and add the four absolute values ​​of difference to obtain the neighborhood difference matrix". Then, the neighborhood difference matrix is ​​linearly normalized and output as the locked boundary map. The calculation process of linear normalization is: "Locked boundary map = (neighborhood difference matrix - minimum value) / (maximum value - minimum value + constant -)". The boundary suppression matrix is ​​obtained by performing fractional mapping on the locked boundary map. The calculation process of fractional mapping is "boundary suppression matrix = constant - / (constant - + locked boundary map)". Then, the boundary suppression matrix is ​​multiplied by the matching confidence map at the pixel level to obtain the constraint matrix and output as the deformation constraint map. The calculation process of pixel-level multiplication is "deformation constraint map = matching confidence map × boundary suppression matrix". Based on the deformation constraint map, the normalized texture map is resampled after treatment to generate a registration texture map. Then, the grayscale difference between the registration texture map and the stable mask map is analyzed to generate an alignment difference base map. The normalized image before treatment is determined from the normalized texture map as the reference image, and the normalized image after treatment is determined as the image to be aligned. The spatial weight of resampling is limited by the deformation constraint map. The horizontal difference matrix and the vertical difference matrix are calculated for the reference image and the image to be aligned, respectively. The square root of the sum of squares is calculated at the same pixel position to obtain the gradient magnitude matrix. Then, the gradient difference matrix is ​​calculated for the two gradient magnitude matrices at the same position. The gradient difference matrix is ​​obtained by taking the absolute difference of the gradient magnitudes at the same position. The gradient difference matrix is ​​multiplied pixel-by-pixel with the deformation constraint map to obtain the constraint difference matrix. The first-order difference of the constraint difference matrix is ​​calculated along the horizontal and vertical directions to obtain the error direction matrix. Then, the error direction matrix is ​​multiplied pixel-by-pixel with the deformation constraint map to obtain the constraint displacement matrix. The constraint displacement matrix is ​​used as the displacement field of the coordinate mapping. Based on the displacement field, coordinate mapping is performed on the image to be aligned, and neighborhood weighted interpolation is used to calculate the resampled pixel value. The calculation process of neighborhood weighted interpolation is "resampled pixel value = Σ(neighborhood pixel value × neighborhood weight) / Σ(neighborhood weight)", where the neighborhood weight is obtained by mapping the distance from the position to be resampled to the neighboring pixel through an exponential function, thereby obtaining the post-treatment registration image. The pre-treatment normalized image and the post-treatment registration image are output as a pair to form the registration texture map. The gray-level difference matrix is ​​obtained by calculating the absolute difference of gray levels at the same pixel position between the pre-treatment normalized image and the post-treatment registered image in the registration texture map. The gray-level difference matrix is ​​obtained by "gray-level difference matrix = |pre-treatment normalized image - post-treatment registered image|". The gray-level difference matrix is ​​multiplied pixel by pixel with the stable mask image to obtain the mask difference matrix. The mask difference matrix is ​​then linearly normalized. The linear normalization calculation process is "normalized difference matrix = (mask difference matrix - minimum value) / (maximum value - minimum value + constant -)". Finally, the normalized difference matrix is ​​output as the alignment difference base map.

[0030] This scheme achieves the suppression of illumination drift and the enhancement of stable texture edges by solving gradients and fusing them pixel-by-pixel with the brightness reference map under a stable mask. The difference of the fusion result is calculated and modulated by a normalized consistency map to obtain a matching confidence map, which reduces weak texture mismatches and outputs a reliable matching distribution. The matching confidence map and the locked boundary map are fused pixel by pixel to form a deformation constraint map, which can limit the deformation of unstable areas and prevent distortion. After resampling and treatment according to the deformation constraint map, a registration texture map is obtained, which improves the alignment of the same area and reduces structural drift. The registration texture map is then subjected to grayscale difference within the stable mask to generate an alignment difference base map: highlighting the real differences and suppressing mismatch pseudo differences.

[0031] The above describes modulating a normalized texture map with a stable mask image after gradient difference calculation to generate an aligned difference base map. The following describes performing multi-order difference weighting on the normalized texture map based on illumination acquisition parameters to generate a texture difference map, specifically including: The parameter amplitude solution results corresponding to the illumination acquisition parameters are subjected to proportional compression mapping with the correction coefficient map to generate a gated base map. Then, a piecewise monotonic transformation is performed on the gated base map to generate a frequency band weight map containing different frequency weights. After performing first-order and second-order differences on the registration texture map corresponding to the normalized texture map, pixel-level weighting is performed on the corresponding frequency weights of the frequency band weight map to generate a response texture map, and the response texture map is output as a texture difference map.

[0032] Among them, the gated base map refers to a pixel-level weighted base image data generated by the illumination acquisition parameters and the correction coefficient map. It is used to apply consistent and interpretable gating constraints to the difference intensity at different spatial locations in subsequent texture difference processing. The frequency band weight map refers to a multi-channel weight data with the same image size, which is calculated based on the illumination acquisition parameters and the correction coefficient map. It is used to describe the relative proportion of different order differential texture responses in different spatial locations when participating in weighted fusion. Response texture map refers to image data generated by performing multi-order difference operations on the registration texture map and combining it with a frequency band weight map for pixel-level weighted fusion.

[0033] The above content will be described in detail below: The solution results of the parameter amplitudes corresponding to the illumination acquisition parameters are subjected to proportional compression mapping with the correction coefficient map to generate a gated base map. The specific calculation formula is as follows: ; In the formula, Indicates the gated base map at the pixel level The gate value at that location, Indicates the correction coefficient map in pixels The coefficient value at that location, This indicates the result of the parameter magnitude calculation. Represents a very small positive number. This indicates the dimension of the illumination acquisition parameters. Indicates the light acquisition parameters at the 1st The values ​​in the dimension are all normalized during the calculations. A piecewise monotonic transformation is performed on the gated base map to generate a frequency band weight map containing weights for different frequencies. The specific calculation formula is as follows: ; ; In the formula, Represents pixels in the frequency band weighting map The frequency band weight vector at that location, This indicates the low-frequency weight at pixel position. The value, Indicates the intermediate frequency weight at pixel position The value, Indicates the high-frequency weights at pixel positions The value, This represents the normalized gate value obtained by normalizing the gated base graph. Indicates the parameters of the first segment. Indicates the second segment parameter, Represents the saturation operator. This represents the minimum value of the entire gated base graph. This represents the maximum value of the entire gated base graph. All the above data have been normalized during the calculation. After performing first-order and second-order differences on the registered texture map corresponding to the normalized texture map, pixel-level weighting is performed on the corresponding frequency weights of the frequency band weight map to generate a response texture map, which is then output as a texture difference map. For first-order differences, for each registered image, first-order difference matrices are calculated in both the horizontal and vertical directions. The horizontal first-order difference matrix is ​​obtained by subtracting the gray levels of adjacent pixels, and the vertical first-order difference matrix is ​​obtained by subtracting the gray levels of adjacent rows of pixels. Then, a first-order difference magnitude matrix is ​​calculated for both matrices at the same pixel location. The calculation process for the first-order difference magnitude matrix is: "First-order difference magnitude = (Horizontal first-order difference...)" 2 + Vertical first-order difference 2 The square root of ) and the amplitude of the second first-order difference = (the first-order difference of the horizontal axis) 2 + Vertical first-order difference 2 "The root of )"; For second-order differences, for each registered image, the second-order difference matrix is ​​calculated in both the horizontal and vertical directions. The horizontal second-order difference matrix is ​​obtained by subtracting two adjacent first-order differences, and the vertical second-order difference matrix is ​​also obtained by subtracting two adjacent first-order differences. Then, the second-order difference magnitude matrix is ​​calculated at the same pixel position for both matrices. The calculation process for the second-order difference magnitude matrix is: "First second-order difference magnitude = (Horizontal second-order difference...)" 2 + Vertical second-order difference 2 The square root of ) and the amplitude of the second second difference = (horizontal second difference) 2 + Vertical second-order difference 2 "The root of )"; Read the intermediate frequency weights corresponding to the first-order difference and the high frequency weights corresponding to the second-order difference from the frequency band weight map, and perform pixel-level weighting to obtain the gated first-order response and the gated second-order response respectively. The calculation process of the gated first-order response is "first gated first-order response = first first-order difference amplitude × intermediate frequency weight, second gated first-order response = second first-order difference amplitude × intermediate frequency weight", and the calculation process of the gated second-order response is "first gated second-order response = first second-order difference amplitude × high frequency weight, second gated second-order response = second second-order difference amplitude × high frequency weight". Pixel-level synthesis is performed on the gated first-order response and gated second-order response to obtain a response texture map, which is then output as a texture difference map. The calculation process of pixel-level synthesis is "first response texture map = gated first-order response + gated second-order response, second response texture map = gated first-order response + gated second-order response". The two response texture maps are linearly normalized according to the minimum and maximum values ​​to ensure that the amplitudes can be compared. This completes the generation and output of the first-order difference and second-order difference weighted response of the registration texture map corresponding to the normalized texture map, and the response texture map is output as a texture difference map.

[0034] This scheme performs proportional compression mapping on the parameter amplitude solution and the correction coefficient map, and generates a frequency band weight map by piecewise monotonic transformation. It constrains the illumination intensity and correction amplitude to a unified gating scale, so that the enhancement intensity under different shooting conditions adaptively converges, reduces noise amplification caused by overexposure gain and suppresses pseudo textures in shadow areas. It performs first-order and second-order difference on the registration texture map and generates a pixel-level weighted texture map according to the frequency band weight map, and outputs the response texture map as a texture difference map. It simultaneously enhances the real structural response of linear valleys and point depressions in the stable same area, weakens the non-structural differences caused by specular reflection and registration residuals, so that the distribution of texture differences before and after is more concentrated, more continuous and more comparable.

[0035] The above describes the generation of a texture difference map by performing multi-level differential weighting on a normalized texture map based on illumination acquisition parameters. The following describes the generation of a response consistency map by combining the texture difference map with an aligned difference base map and performing fractional mapping, specifically including: Perform co-domain linear normalization and pixel-by-pixel fusion on the alignment difference base map and deformation constraint map respectively to generate a structural confidence map; The absolute difference is calculated on the texture difference map, and the result of the absolute difference calculation is weighted and normalized pixel by pixel and fractionally mapped according to the structure confidence map to generate a response consistency map.

[0036] Among them, the structural confidence map refers to image-type intermediate data generated within the same pixel coordinate domain based on the numerical relationship between the alignment difference base map and the deformation constraint map, which is used to represent the distribution of the structural reliability of the texture structure at each pixel location under spatial alignment and deformation constraint conditions. The absolute difference calculation result refers to the image data obtained by performing a pixel-by-pixel absolute difference operation on the response texture maps corresponding to before and after treatment at the same spatial coordinate position.

[0037] The above content will be described in detail below: Perform in-domain linear normalization and pixel-by-pixel fusion on the alignment difference base map and deformation constraint map respectively to generate a structure confidence map: The minimum and maximum gray values ​​of the entire image are obtained from the alignment difference base map. Then, a normalization calculation is performed on each pixel: "Normalized difference value = (difference pixel value - minimum gray value) / (maximum gray value - minimum gray value + constant -)" to obtain the normalized difference map. Then, the normalized difference map is reverse mapped to obtain the difference suppression map. The calculation process of the reverse mapping is: "Difference suppression value = constant - / (constant - + normalized difference value)". The minimum and maximum gray values ​​of the entire image are obtained from the deformation constraint map, and a normalization calculation is performed on each pixel: "Normalization constraint value = (constraint pixel value - minimum gray value) / (maximum gray value - minimum gray value + constant -)" to obtain the normalized constraint map; At the same coordinate position, pixel-wise multiplication fusion is performed on the difference suppression map and the normalization constraint map. The fusion calculation process is "structure confidence value = difference suppression value × normalization constraint value". Then, the obtained structure confidence value matrix is ​​subjected to linear normalization calculation again "structure confidence value' = (structure confidence value - structure confidence minimum value) / (structure confidence maximum value - structure confidence minimum value + constant -)" to unify the output scale. The final output is the structure confidence map as the image matrix. The absolute difference of the texture difference map is calculated, and the result of the absolute difference calculation is then weighted and normalized pixel-wise according to the structure confidence map to generate a response consistency map. The absolute difference is calculated on the texture difference map to obtain the absolute difference calculation result. The calculation process is as follows: take the absolute value of the difference value at any pixel position in the texture difference map to obtain the absolute difference value at the corresponding pixel position and form the absolute difference calculation result. The absolute difference calculation result is weighted and normalized pixel by pixel using the structure confidence map to obtain a weighted normalized difference map. The calculation process is as follows: For each pixel position, calculate the weighted difference value = absolute difference calculation result × structure confidence value, and sum the weighted difference values ​​of all pixels to obtain the total weighted difference. At the same time, sum the structure confidence values ​​of all pixels to obtain the total confidence weight. Then calculate the weighted mean difference = total weighted difference / (total confidence weight + constant). Finally, calculate the normalized difference value for each pixel position = weighted difference value / (weighted mean difference + constant), thus obtaining the weighted normalized difference map. A fractional mapping is performed on the weighted normalized difference map to generate a response consistency map. The calculation process is as follows: calculate the consistency value for each pixel position = constant - / (constant - + normalized difference value), and output the response consistency map by combining all consistency values.

[0038] This scheme performs co-domain linear normalization on the alignment difference base map and deformation constraint map, and then fuses them pixel by pixel to generate a structural confidence map. It unifies the registration stability and difference perturbation intensity into confidence weights of the same scale, thereby spatially suppressing the influence of mismatched regions and high-perturbation regions and highlighting comparable regions. It performs absolute difference calculation on the texture difference map and performs pixel-by-pixel weighted normalization according to the structural confidence map, and then performs fractional mapping to generate a response consistency map. This makes the consistency measure mainly determined by the real texture changes in high-confidence regions, reducing the artificial height difference caused by noise and residual illumination. Furthermore, the fractional mapping separates the consistency levels to improve the stability and interpretability of the before and after response comparison.

[0039] The above describes performing fractional mapping on the texture difference map combined with the alignment difference base map to generate a response consistency map. The following describes performing consistent gating encoding on the response consistency map to generate a structure index map, specifically including: The texture difference map and the structure confidence map are weighted to generate a gated response. The gated response is then calculated by difference in the horizontal and vertical directions and compared with the matching confidence map to extract the sign change position and perform binarization to generate a wrinkle skeleton map. Based on the gating response and response consistency map, the wrinkle skeleton map is subjected to dual-time consistency screening to generate pore candidate maps. The wrinkle skeleton map and pore candidate maps are then synthesized through mutual exclusion and superimposed with the registered texture map to generate a structure index map.

[0040] Among them, gated response refers to the image data obtained by weighted fusion of texture difference map and structure confidence map pixel by pixel at the same pixel coordinate position; Wrinkle skeleton map refers to binary image data that represents the direction of wrinkle center and the position of continuous structure on the skin surface, calculated by directional difference and sign change extraction methods based on texture difference map, structure confidence map and matching confidence map during the consistency constraint analysis of skin texture changes before and after treatment. Pore ​​candidate images refer to spatial distribution image data that may belong to pore structures, obtained by combining gated response and response consistency maps with an existing wrinkle skeleton image to screen for consistency differences in non-linear, dotted, or local clustered textures.

[0041] The above content will be described in detail below: The texture difference map and the structure confidence map are weighted to generate a gated response. The gated response is then calculated by subtracting from it in the horizontal and vertical directions, and the sign change position is extracted and binarized with the matching confidence map to generate a wrinkle skeleton map. The texture difference map and the structure confidence map are weighted to generate a gated response. The weighting calculation process is as follows: for each pixel position, the gated response value is calculated as texture difference value × structure confidence value, and all gated response values ​​are linearly normalized according to the minimum and maximum values ​​to obtain the gated response. The gating response is calculated using both horizontal and vertical differential methods. The horizontal differential method is calculated as follows: for each pixel position, the horizontal differential value is calculated as the difference between the gating response value at that pixel position and the pixel position to its right. The vertical differential method is calculated as follows: for each pixel position, the vertical differential value is calculated as the difference between the gating response value at that pixel position and the pixel position to its right. After obtaining the horizontal and vertical difference maps, they are fused to form a difference magnitude map. The calculation process for the difference magnitude map is as follows: For each pixel position, calculate the difference magnitude = √(horizontal difference value). 2 +Vertical difference value 2 The difference magnitude map is then linearly normalized to its minimum and maximum values ​​to obtain a normalized difference magnitude map. The normalized difference magnitude map and the matching confidence map are jointly constrained to suppress fractures caused by mismatches. The calculation process of the joint constraint is as follows: For each pixel position, calculate the constraint difference value = normalized difference magnitude × matching confidence value, and then linearly normalize the constraint difference value according to the minimum and maximum values ​​to obtain the constraint difference map. The sign change position extraction is performed on the constraint difference map to determine the candidate position of the wrinkle center line. The calculation process of the sign change position extraction is as follows: Calculate the sign product of adjacent differences in the horizontal and vertical directions of the constraint difference map respectively. If the sign product of the differences of two adjacent points in the same direction is less than zero, it is determined that there is a sign change at that position. The union of the sign change positions in the horizontal direction and the sign change positions in the vertical direction is taken to obtain the sign change set map. The symbol change set map is binarized to output a wrinkle skeleton map. The binarization process is as follows: the position in the symbol change set map that belongs to the symbol change set is assigned a value of one, and the other positions are assigned a value of zero, so as to obtain a wrinkle skeleton map represented by a pixel-level binary matrix. Based on the gating response and response consistency map, a two-timetime consistency screening is performed on the wrinkle skeleton map to generate pore candidate maps. The wrinkle skeleton map and pore candidate maps are then synthesized through mutual exclusion and superimposed on the registered texture map for encoding to generate a structure index map. The gating response is fused at two time points to form a consistent gating intensity map. The calculation process is as follows: the gating response before treatment and the gating response after treatment are averaged at the same pixel position to obtain the gating mean map, that is, gating mean map = (gating response before treatment + gating response after treatment) / 2. The gating mean map and the response consistency map are then multiplied pixel by pixel to obtain the consistent gating intensity map, that is, consistent gating intensity map = gating mean map × response consistency map. A two-timetime consistency screening is performed on the wrinkle skeleton map to generate a screened skeleton map. The calculation process is as follows: the wrinkle skeleton map is multiplied pixel by pixel by the consistency gate intensity map to obtain the skeleton consistency weight map, that is, skeleton consistency weight map = wrinkle skeleton map × consistency gate intensity map. Then, the skeleton consistency weight map is linearly normalized to obtain a normalized skeleton weight map. The linear normalization adopts the formula: normalized skeleton weight map = (skeleton consistency weight map - minimum value) / (maximum value - minimum value + constant -). Subsequently, the normalized skeleton weight map is fractionally mapped to obtain the screened skeleton map. The fractional mapping adopts the formula: screened skeleton map = normalized skeleton weight map / (normalized skeleton weight map + constant -). The process of generating pore candidate images from the skeleton image involves several independent data processing steps. First, a local neighborhood mean map is calculated for the consistent gating intensity map. This local neighborhood mean map is obtained by performing an arithmetic mean on the consistent gating intensity map within a fixed window. Then, a point difference amplitude map is calculated: Point difference amplitude map = (Local neighborhood mean map - Consistent gating intensity map). 2 The point difference amplitude map is linearly normalized to obtain a normalized point difference map. Then, in order to suppress the influence of linear structure on point candidates, the normalized point difference map and the screening skeleton map are complementaryly gated and the pore candidate map is output. The complementary gating adopts the formula: pore candidate map = normalized point difference map × [constant - / (constant - + screening skeleton map)], so that the pore candidate value is jointly determined by the point difference amplitude and the complementary weight of the skeleton. The selected skeleton image and pore candidate image are subjected to mutual exclusion synthesis to form mutually exclusive skeleton images and mutually exclusive pore images. The independent data processing process is to first calculate the mutual exclusion weight map, mutual exclusion weight map = constant 1 / (constant 1 + selected skeleton image + pore candidate image), and then calculate the mutual exclusion skeleton image = selected skeleton image × mutual exclusion weight map and the mutually exclusive pore image = pore candidate image × mutual exclusion weight map respectively. The mutually exclusive skeleton image and mutually exclusive pore image are synthesized by mutual exclusion and then superimposed on the registration texture image to generate a structure index image. The independent data processing process is as follows: first, the registration image used for superposition in the registration texture image is linearly normalized to obtain a normalized registration image; then, the index value image is calculated as: Index value image = Normalized registration image × Constant 2 + Mutually exclusive skeleton image × Constant 3 + Mutually exclusive pore image × Constant 4. The index value image is then linearly normalized to obtain the structure index image. The linear normalization method is: Structure index image = (Index value image - Minimum value) / (Maximum value - Minimum value + Constant 1).

[0042] This scheme generates a gated response by weighting the texture difference map and the structure confidence map: this concentrates the difference energy in the high-confidence region, suppresses pseudo-differences caused by reflection and weak texture noise, performs horizontal / vertical difference on the gated response and combines it with the matching confidence map to extract the sign change position and binarize it to generate a wrinkle skeleton map, enhancing the continuity and positioning consistency of linear valleys and reducing breaks and false connections caused by mismatches. Based on the gated response and the response consistency map, the wrinkle skeleton map is subjected to dual-timetime consistency screening to generate pore candidate maps, eliminating pseudo-point structures that only appear at a single time and improving the temporal stability of pore candidates. The wrinkle skeleton map and pore candidate maps are mutually exclusive and synthesized and superimposed with the registered texture map to generate a structure index map, avoiding mutual encroachment of structures and achieving unified indexing of structures under the same coordinate area, which facilitates verification and comparison.

[0043] The above describes the consistent gating encoding of the response consistency graph to generate a structure index graph. The following describes using the structure index graph as a structural constraint to perform difference enhancement on the alignment difference base map, generating an improved saliency map, specifically including: Perform piecewise monotonic fractional mapping on the structure index map to generate a structure constraint matrix, and perform pixel-wise multiplication and fusion on the absolute difference calculation result and the structure constraint matrix to generate a structure constraint difference matrix. Perform boundary-suppression fractional mapping on the aligned difference base map to generate a difference-enhancing weight matrix; The structural constraint difference matrix and the difference enhancement weight matrix are fused by pixel-by-pixel multiplication and extreme value interval clipping to generate an improved saliency map.

[0044] The structural constraint matrix refers to a two-dimensional numerical matrix generated from the structural index map using deterministic numerical mapping rules, where the pixel dimension and alignment difference are... Figure 1 This is used to perform structural location constraints and weight modulation on differential responses in improving significance calculations; The difference enhancement weight matrix refers to weighted image data that is directly calculated based on the alignment difference base map and has the same spatial size and pixel correspondence as the alignment difference base map.

[0045] The above content will be described in detail below: Perform piecewise monotonic fractional mapping on the structure index graph to generate the structure constraint matrix: The average index value and the maximum index value of the entire structure index map are calculated. The average index value is obtained by summing all pixel values ​​and dividing by the total number of pixels, while the maximum index value is obtained by taking the maximum of all pixel values. Then, using the average index value as the segmentation point, a piecewise fractional mapping is performed pixel-by-pixel on the structure index map, ensuring that the mapping function is monotonic within each segment and continuous overall. The specific calculation process is as follows: when a pixel index value is not greater than the average index value of the entire image, the lower segment constraint value is calculated as: = pixel index value / (average index value of the entire image + constant -); when... When a pixel index value is greater than the average index value of the entire image, first calculate the high-segment normalization difference = (pixel index value - average index value of the entire image) / (maximum index value of the entire image - average index value of the entire image + constant 1), then calculate the high-segment constraint value = average index value of the entire image / (average index value of the entire image + constant 1) + high-segment normalization difference / (constant 1 + high-segment normalization difference). This ensures that the low-segment constraint value increases with the increase of the pixel index value without exceeding the upper limit of the first segment, and that the high-segment constraint value continues to increase based on the upper limit of the first segment with the increase of the high-segment normalization difference without exceeding the constant 1. The constraint values ​​obtained by mapping each pixel position according to the above piecewise fraction are used to form a matrix of the same size as the structure index map and output as the structure constraint matrix. The absolute difference calculation result is fused with the structural constraint matrix by performing pixel-by-pixel multiplication to generate the structural constraint difference matrix; Perform boundary suppression fractional mapping on the aligned difference basemap to generate a difference enhancement weight matrix: The pixel matrix of the aligned difference base map is denoted as the difference value matrix. The minimum and maximum difference values ​​of the difference value matrix are calculated. Then, linear normalization is performed on the difference value matrix to obtain the normalized difference matrix. The normalization calculation process is "normalized difference matrix = (difference value matrix - minimum difference value) / (maximum difference value - minimum difference value + constant -)" so that the value of the normalized difference matrix falls within the range of zero to one. To achieve boundary suppression, the boundary magnitude matrix is ​​calculated from the normalized difference matrix. The calculation process of the boundary magnitude matrix is ​​as follows: the first-order difference of the normalized difference matrix is ​​calculated in the horizontal and vertical directions to obtain the horizontal difference matrix and the vertical difference matrix, respectively. Then, the boundary magnitude matrix is ​​calculated at the same pixel position as the square root of the horizontal difference matrix plus the square of the vertical difference matrix. This allows the boundary magnitude matrix to characterize the degree of spatial abrupt change in the difference distribution. The normalized difference matrix and the boundary magnitude matrix are fused to obtain the boundary suppression difference matrix. The fusion calculation process is "boundary suppression difference matrix = normalized difference matrix × constant - / (constant - + boundary magnitude matrix)" to reduce the difference contribution at the boundary abrupt change position. A fractional mapping is performed on the boundary suppression difference matrix to generate a difference enhancement weight matrix. The calculation process of the fractional mapping is "difference enhancement weight matrix = constant - / (constant - + boundary suppression difference matrix)". The difference enhancement weight matrix is ​​then linearly normalized according to the minimum and maximum values, so that the positions with smaller differences and weaker boundary abruptness receive larger weight values, and the positions with larger differences or stronger boundary abruptness receive smaller weight values, thus obtaining the difference enhancement weight matrix. The structural constraint difference matrix and the difference enhancement weight matrix are fused through pixel-by-pixel multiplication and extreme value interval clipping to generate an improved saliency map: Read the difference value of the structural constraint difference matrix and the weight value of the difference enhancement weight matrix at the same pixel coordinate position. First, calculate the fusion matrix, where the calculation process of the fusion matrix is ​​"fusion matrix = structural constraint difference matrix × difference enhancement weight matrix", so that the difference value is enhanced or suppressed under the action of the weight. The extreme value interval clipping of the fusion matrix is ​​used to obtain the clipping matrix. The calculation process of the clipping matrix is ​​as follows: calculate the global minimum and global maximum values ​​of the fusion matrix, and linearly determine the lower clipping limit and upper clipping limit using the global minimum and global maximum values. The calculation process of the lower clipping limit is "lower clipping limit = global minimum + (global maximum - global minimum) / constant 4", and the calculation process of the upper clipping limit is "upper clipping limit = global maximum - (global maximum - global minimum) / constant 4". Then, the fusion value of each pixel in the fusion matrix is ​​subject to interval restriction. The calculation process of the interval restriction is "when the fusion value is less than the lower clipping limit, take the lower clipping limit; when the fusion value is greater than the upper clipping limit, take the upper clipping limit; otherwise, take the original fusion value", thereby suppressing the interference of outliers with abnormally bright and abnormally dark values ​​on the saliency distribution. The magnitude normalization of the clipping matrix is ​​performed to output the improved saliency map. The normalization calculation process is "Improved saliency map = (Clipping matrix - Lower clipping limit) / (Upper clipping limit - Lower clipping limit + Constant)".

[0046] This scheme performs piecewise monotonic fractional mapping on the structure index map to generate a structure constraint matrix, and then fuses it with the absolute difference result pixel-by-pixel multiplication to generate a structure constraint difference matrix. This limits the difference response to the neighborhood of real structures such as wrinkles and pores and suppresses random differences in non-structure regions, reducing the contribution of noise to saliency from the source. Boundary suppression fractional mapping is performed on the alignment difference base map to generate a difference enhancement weight matrix. Weight attenuation is applied to areas with high alignment differences and sharp boundary transitions to reduce the amplification of spurious differences introduced by mismatches and deformed boundaries. The structure constraint difference matrix and the difference enhancement weight matrix are fused pixel-by-pixel multiplication and extreme value intervals are cropped to generate an improved saliency map. While maintaining the concentration of saliency space and controllable amplitude, this scheme improves the contrast and interpretability of the real improvement area and avoids a small number of abnormal differences dominating the overall saliency result.

[0047] The above describes using the structural index map as a structural constraint to perform difference enhancement on the alignment difference base map and generate an improved saliency map. The following describes using the structural index map as a structural constraint to perform difference enhancement on the alignment difference base map and generate an improved saliency map, specifically including: The structural confidence map and the improvement saliency map are multiplied pixel by pixel to generate an improvement heatmap. The improvement heatmap is then projected onto the wrinkle skeleton map and the pore candidate map and weighted according to the frequency band weight map to generate a score overlay map. Within the locked boundaries, the score overlay map and the improvement significance map are analyzed for differences and then overlaid with the response consistency map to generate a review difference map. The registered texture map, structural index map, scoring overlay map, and verification difference map are stitched together to generate an analysis report.

[0048] Among them, improved heatmaps refer to image data used to characterize the spatial distribution of the intensity of structural-related texture changes in the same treatment area before and after treatment; The scoring overlay image refers to two-dimensional image data formed by multi-source mapping and fusion of the improvement saliency map under structural and frequency band weight constraints in the same pixel coordinate system. The verification difference map refers to the image data obtained by calculating the spatial consistency difference between the score overlay map and the improvement saliency map at the pixel level within the effective analysis area defined by the locked boundary, and then combining it with the response consistency map for consistency modulation.

[0049] The above content will be described in detail below: The structural confidence map and the improvement saliency map are multiplied pixel by pixel to generate an improvement heatmap; After projecting the improved heatmap onto the wrinkle skeleton map and pore candidate map, the scores are weighted according to the frequency band weight map to generate a score overlay map. The improved heatmap is projected onto the wrinkle skeleton map to obtain the wrinkle projection map. The calculation process is "wrinkle projection map = improved heatmap × wrinkle skeleton map", where the multiplication is performed pixel by pixel, so that the improved heatmap is retained only in the pixel positions covered by the wrinkle skeleton map. The improved heatmap is projected onto the pore candidate map to obtain the pore projection map. The calculation process is "pore projection map = improved heatmap × pore candidate map", so that the improved heatmap is retained only in the pixel positions covered by the pore candidate map. The wrinkle projection map and pore projection map are weighted according to the frequency band weight map to obtain the frequency band weighted wrinkle map and the frequency band weighted pore map respectively. The calculation process of the frequency band weighted wrinkle map is "frequency band weighted wrinkle map = wrinkle projection map × high frequency weight in the frequency band weight map", and the calculation process of the frequency band weighted pore map is "frequency band weighted pore map = pore projection map × mid frequency weight in the frequency band weight map", so that the improvement intensity of different structures under different frequency band weights is modulated differently. The frequency band-weighted wrinkle image and the frequency band-weighted pore image are combined to output a score overlay image. The calculation process of the combination is "score overlay image = (frequency band-weighted wrinkle image + frequency band-weighted pore image) / 2". The score overlay image is then linearly normalized to map its pixel values ​​to a uniform display range, thereby obtaining a score overlay image based on frequency band weights in the wrinkle and pore structure domains. Within the locked boundaries, a difference analysis is performed on the score overlay plot and the improvement significance plot, and then overlaid with the response consistency plot to generate a review difference plot: The magnitude of the locked boundary map is normalized to obtain a normalized boundary map, and the normalized boundary map is then subjected to fractional mapping to generate an effective domain map. The calculation process of fractional mapping is "effective domain map = 1 / (1+normalized boundary map)", which makes the effective domain weight smaller at the location with stronger boundary transition. The difference matrix is ​​obtained by calculating the absolute difference between the score overlay map and the improvement saliency map at the same pixel position. The calculation process of the difference matrix is ​​"difference matrix = |score overlay map - improvement saliency map|". Then, the difference matrix is ​​multiplied pixel by pixel with the effective domain map to obtain the intra-domain difference map. The calculation process of the intra-domain difference map is "intra-domain difference map = difference matrix × effective domain map". The intra-domain difference map and the response consistency map are overlaid and fused to generate a verification difference map. The data processing method is as follows: first, the amplitude of the response consistency map is normalized to obtain a normalized consistency map, and then a complementary transformation is performed on the normalized consistency map to obtain a consistency suppression map. The calculation process of the consistency suppression map is "consistency suppression map = constant - normalized consistency map". Subsequently, the intra-domain difference map and the consistency suppression map are multiplied pixel by pixel to obtain a verification mark map. The calculation process of the verification mark map is "verification mark map = intra-domain difference map × consistency suppression map". Finally, the verification mark map is linearly normalized according to the minimum and maximum values ​​to output the verification difference map. The registered texture map, structural index map, scoring overlay map, and verification difference map are stitched together to generate an analysis report. First, size uniformity processing is performed on the registration texture map, structure index map, scoring overlay map, and verification difference map respectively. The size uniformity calculation process is to calculate the width and height of each image and determine the target width and target height. The target width is the maximum value of the width of the four images, and the target height is the maximum value of the height of the four images. Then, boundary padding is performed on any image whose width is less than the target width or whose height is less than the target height to obtain a padded image. The boundary padding calculation process is to pad the right side of the image with a blank column corresponding to the difference between the target width and the current width, and to pad the bottom of the image with a blank row corresponding to the difference between the target height and the current height. The pixel values ​​of the blank column and blank row are taken as the average value of all pixels in the image. After obtaining four completed images, grayscale range unification is performed on each of the four completed images. The calculation process for grayscale range unification is to calculate the minimum and maximum pixel values ​​for each completed image and perform linear normalization. The linear normalization result is calculated as "normalized pixel = (pixel - minimum value) / (maximum value - minimum value + constant -)". The process involves constructing a spliced ​​canvas, completing the layout copy, and generating an analysis report. The spliced ​​canvas generation process involves determining the splicing method as a two-row, two-column layout and calculating the canvas width and height. The canvas width is equal to the target width multiplied by a constant two, and the canvas height is equal to the target height multiplied by a constant two. The machine initializes the canvas pixel matrix and copies four normalized images to the four quadrant regions of the canvas. The copying calculation process involves assigning the pixel value of each pixel position of each normalized image to the corresponding offset pixel unit on the canvas. Specifically, the registration texture map is copied to the upper left quadrant, the structure index map is copied to the upper right quadrant, the scoring overlay map is copied to the lower left quadrant, and the verification difference map is copied to the lower right quadrant. To ensure clear stitching boundaries, quadrant separator lines are calculated and written to separator line pixels. The calculation process for separator line pixels involves taking the average of the pixel values ​​of four normalized images as the line value and assigning the line value to the position on the canvas where the row index equals the target height and the column index equals the target width. The overall amplitude is renormalized on the spliced ​​canvas. The calculation process of overall amplitude renormalization is to calculate the minimum and maximum values ​​of the canvas and perform linear normalization according to "output pixel = (canvas pixel - minimum value) / (maximum value - minimum value + constant -)", thereby outputting an analysis report containing the splicing results of the registration texture map, structure index map, scoring overlay map and verification difference map.

[0050] This solution generates an improvement heatmap by multiplying the structural confidence map and the improvement saliency map pixel by pixel. This can suppress false highlights in low-confidence areas while preserving true changes in high-confidence structures, ensuring that the heatmap distribution is consistent with reliable texture locations. The improvement heatmap is projected onto the wrinkle skeleton map and pore candidate map and weighted according to the frequency band weight map to generate a score overlay map. This can limit the improvement intensity to the wrinkle and pore structural domains and assign weights according to frequency band contributions, making the comprehensive score correspond to the structural category and avoiding interference from non-structural textures. Within the locked boundary, the difference analysis between the score overlay map and the improvement saliency map is performed and overlaid with the response consistency map to generate a verification difference map. This can locate areas where the score and significant changes are inconsistent and use consistency information to suppress occasional noise, thereby improving the readability of the verification. The registered texture map, structural index map, score overlay map, and verification difference map are stitched together to generate an analysis report, which can simultaneously present the aligned base map, structural location, score distribution, and verification markers in the same coordinate system, facilitating rapid comparison and traceability.

[0051] Example 2: Please see Figure 5A texture image analysis system for skin rejuvenation treatment effects, including: The data acquisition module is used to acquire the treatment texture image and lighting data of the target object; The stability analysis module is used to perform separation operations on the treatment texture image according to the treatment sequence, and to evaluate the symmetry consistency between the illumination acquisition data and the separation operation results to generate a stable mask image. The brightness analysis module is used to perform weighted alignment of the separation operation results before and after treatment based on the stable mask image, generate locked texture images in pairs, and perform symmetric statistics on the locked texture images to generate a brightness reference image. The difference analysis module is used to perform fusion correction on the brightness reference map and the locked texture map according to the illumination acquisition parameters, generate a normalized texture map, and modulate the normalized texture map with the stable mask map after gradient difference calculation to generate an aligned difference base map. The response analysis module is used to perform multi-level differential weighting on the normalized texture map based on the illumination acquisition parameters to generate a texture difference map, and to perform fractional mapping on the texture difference map in combination with the alignment difference base map to generate a response consistency map. The analysis report generation module is used to perform consistent gating encoding on the response consistency graph and generate a structure index graph. Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map. The improved saliency map and the normalized texture map are then image-stitched together to generate an analysis report.

[0052] This embodiment has the same technical effects as Embodiment 1.

[0053] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. The data mentioned in this application, when used for calculations, have undergone normalization and other preprocessing to achieve dimensional uniformity.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing the texture image of skin rejuvenation treatment effects, characterized in that, Includes the following steps: Acquire the treatment texture image and lighting acquisition data of the target object; The treatment texture image is subjected to a separation operation according to the treatment time sequence, and the illumination acquisition data and the separation operation results are evaluated for symmetry consistency to generate a stable mask image. The separation operation results are weighted and aligned before and after treatment based on the stable mask image to generate a locked texture image, and symmetric statistics are performed on the locked texture image to generate a brightness reference image. The brightness reference map and the locked texture map are fused and corrected according to the illumination acquisition parameters to generate a normalized texture map. The normalized texture map is then modulated with the stable mask map after gradient difference calculation to generate an aligned difference base map. The normalized texture map is subjected to multi-order differential weighting based on the illumination acquisition parameters to generate a texture difference map, and the texture difference map is combined with the alignment difference base map to perform fractional mapping to generate a response consistency map. The response consistency graph is subjected to consistent gating encoding to generate a structure index graph; Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map, and the improved saliency map and the normalized texture map are image-stitched together to generate an analysis report; Specifically, generating a stable mask image involves performing gray-level difference analysis on the separation operation results before and after treatment, pixel by pixel. The illumination acquisition parameters are solved for parameter amplitude, and the grayscale difference analysis results are normalized by mean and fractional mapping based on the parameter amplitude solution results to generate a stable mask image.

2. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 1, characterized in that: The process of performing fusion correction on the brightness reference map and the locked texture map based on the illumination acquisition parameters to generate a normalized texture map specifically includes: Within the stable mask image, the parameter amplitude solution results corresponding to the illumination acquisition parameters are gated and fused with the brightness reference image to generate a correction coefficient image; The correction coefficient map is multiplied pixel by pixel with the locked texture map to generate a coefficient modulation map, and the coefficient modulation map is then fused pixel by pixel with the brightness reference map to generate a normalized texture map.

3. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 2, characterized in that: The process of modulating the normalized texture map with the stable mask map after gradient difference calculation to generate an aligned difference base map specifically includes: Gradient calculations are performed on the normalized texture map in both the horizontal and vertical directions, and the gradient calculation results are then fused pixel-by-pixel with the brightness reference map under the stable mask map. The pixel-by-pixel fusion results are subtracted and modulated by a normalized consensus map to obtain a matching confidence map. The matching confidence map and the locked boundary map are then fused pixel by pixel to generate a deformation constraint map. The normalized consistent map is obtained by performing differential mapping on the normalized texture map; The locked boundary map is obtained by performing neighborhood difference on the stable mask map; The normalized texture map is resampled after treatment based on the deformation constraint map to generate a registration texture map. The grayscale difference between the registration texture map and the stable mask map is analyzed to generate an alignment difference base map.

4. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 3, characterized in that: The process of generating a texture difference map by performing multi-level difference weighting on the normalized texture map based on the illumination acquisition parameters specifically includes: The solution results of the parameter amplitude corresponding to the illumination acquisition parameters are subjected to proportional compression mapping with the correction coefficient map to generate a gated base map, and a piecewise monotonic transformation is performed on the gated base map to generate a frequency band weight map containing different frequency weights; After performing first-order and second-order differences on the registration texture map corresponding to the normalized texture map, pixel-level weighting is performed on the corresponding frequency weights of the frequency band weight map to generate a response texture map, and the response texture map is output as a texture difference map.

5. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 4, characterized in that: Performing a fractional mapping on the texture difference map and the alignment difference base map to generate a response consistency map specifically includes: Perform co-domain linear normalization and pixel-by-pixel fusion on the alignment difference base map and the deformation constraint map respectively to generate a structure confidence map; The absolute difference is calculated on the texture difference map, and the absolute difference calculation result is weighted and normalized pixel by pixel and fractionally mapped according to the structure confidence map to generate a response consistency map.

6. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 5, characterized in that: The process of performing consistent gating encoding on the response consistency graph to generate a structure index graph specifically includes: The texture difference map and the structure confidence map are weighted to generate a gated response. The gated response is then calculated by difference in the horizontal and vertical directions and compared with the matching confidence map to extract the sign change position and perform binarization to generate a wrinkle skeleton map. Based on the gating response and the response consistency map, the wrinkle skeleton map is subjected to dual-time consistency screening to generate a pore candidate map. The wrinkle skeleton map and the pore candidate map are then synthesized through mutual exclusion and superimposed and encoded with the registered texture map to generate a structure index map.

7. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 5, characterized in that: Using the structural index map as a structural constraint, performing difference enhancement on the alignment difference base map to generate an improved saliency map specifically includes: A piecewise monotonic fractional mapping is performed on the structure index map to generate a structure constraint matrix, and the absolute difference calculation result is fused with the structure constraint matrix by pixel-wise multiplication to generate a structure constraint difference matrix. Perform boundary suppression fractional mapping on the aligned difference base map to generate a difference enhancement weight matrix; The structural constraint difference matrix and the difference enhancement weight matrix are multiplied and fused pixel by pixel and the extreme value interval is clipped to generate an improved saliency map.

8. The method for analyzing the texture image of skin rejuvenation treatment effect according to claim 6, characterized in that: The analysis report is generated by image-stitching the improved saliency map and the normalized texture map, specifically including: The structural confidence map and the improvement saliency map are multiplied pixel by pixel to generate an improvement heatmap. The improvement heatmap is then projected onto the wrinkle skeleton map and the pore candidate map and weighted according to the frequency band weight map to generate a score overlay map. Within the locked boundary, the difference analysis of the score overlay map and the improvement significance map is performed and then overlaid with the response consistency map to generate a verification difference map; The registration texture map, the structure index map, the scoring overlay map, and the verification difference map are stitched together to generate an analysis report.

9. A texture image analysis system for skin rejuvenation treatment effects, characterized in that, include: The data acquisition module is used to acquire the treatment texture image and lighting data of the target object; The stability analysis module is used to perform separation operations on the treatment texture image according to the treatment time sequence, and to perform symmetric consistency evaluation on the illumination acquisition data and the separation operation results to generate a stable mask image; The brightness analysis module is used to perform weighted alignment of the separation operation results before and after treatment based on the stable mask image, generate locked texture images in pairs, and perform symmetric statistics on the locked texture images to generate a brightness reference image. The difference analysis module is used to perform fusion correction on the brightness reference map and the locked texture map according to the illumination acquisition parameters, generate a normalized texture map, and modulate the normalized texture map with the stable mask map after gradient difference calculation to generate an aligned difference base map. The response analysis module is used to perform multi-order differential weighting on the normalized texture map according to the illumination acquisition parameters to generate a texture difference map, and to perform fractional mapping on the texture difference map in combination with the alignment difference base map to generate a response consistency map. The analysis report generation module is used to perform consistent gating encoding on the response consistency graph to generate a structure index graph; Using the structural index map as a structural constraint, difference enhancement is performed on the alignment difference base map to generate an improved saliency map, and the improved saliency map and the normalized texture map are image-stitched together to generate an analysis report; Specifically, generating a stable mask image involves performing gray-level difference analysis on the separation operation results before and after treatment, pixel by pixel. The illumination acquisition parameters are solved for parameter amplitude, and the grayscale difference analysis results are normalized by mean and fractional mapping based on the parameter amplitude solution results to generate a stable mask image.

Images