A method for testing performance of clothing fabrics
Through superpixel segmentation and gradient image analysis, combined with multiple denoising processing, the problem of poor denoising effect in hyperspectral imaging technology is solved, and the accuracy of image quality and color fastness detection is improved.
Patent Information
- Application Number
- CN202411110060.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-14
AI Technical Summary
During the denoising process of existing hyperspectral imaging technology, the initial denoising effect is incomplete, and the secondary overall denoising is severely smooth, resulting in low image quality and affecting the accuracy of color fastness detection.
The band image is obtained through superpixel segmentation, the blocked area is obtained, the noise possibility is calculated, and multiple denoising processing is performed to ensure the accuracy of the noise level in each area of the image.
It improves the image noise removal effect, enhances the reliability and accuracy of color fastness detection, and ensures the reliability of experimental results of clothing fabric performance tests.
Smart Images

Figure CN118644481B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of optical detection, and in particular to a method for testing the performance of clothing fabrics. Background Art
[0002] The spectrophotometric method most commonly used in industry to measure fabric color can only measure monochrome fabrics. It has high requirements on fabric size and is cumbersome to operate, making it unsuitable for measuring the color of multi-color fabrics or single yarns.
[0003] In order to meet the needs of enterprise production and development, our company has conducted in-depth research on the precise measurement method of fabric color. In view of the shortcomings of current technology, a hyperspectral imaging system is established based on the analysis of spectral imaging technology, and a fabric color measurement method based on hyperspectral imaging technology is proposed, which realizes color measurement with high accuracy. Spectral imaging technology combines spectral technology and imaging technology together to measure the reflectivity of spectral bands in the region of interest (ROI) of fabrics. In this way, more textile color information can be obtained and higher color measurement accuracy can be achieved.
[0004] The images acquired using a spectrometer may contain noise, which will affect the quality of the spectral image to varying degrees. Therefore, when acquiring images using a spectrometer, the noise needs to be properly processed and removed to improve the accuracy and reliability of the image. Summary of the invention
[0005] The present invention provides a method for testing the performance of clothing fabrics to solve the problem that the primary denoising effect is incomplete and the secondary overall denoising is seriously smooth. The technical scheme adopted is as follows:
[0006] An embodiment of the present invention provides a method for testing the performance of clothing fabrics, the method comprising the following steps:
[0007] Acquire spectral images;
[0008] The spectral image is segmented by superpixel to obtain several superpixel blocks, several band images of the spectral image are obtained, the band images are clustered to obtain block areas for the superpixel blocks, similar superpixel blocks of the superpixel blocks are obtained, several pixels with the largest differences are obtained according to the superpixel blocks and similar superpixel blocks, and the possibility of noise pixels appearing in each superpixel block is obtained according to the several pixels with the largest differences and the number of similar superpixel blocks;
[0009] Obtain the co-located area of the block area, obtain the contour similarity of the block area and the co-located area, obtain the difference value according to the grayscale value of each superpixel block in the block area, obtain the similarity of the block area and the co-located area according to the difference value and the contour similarity, obtain the grayscale area difference according to the grayscale difference between each block area and the superpixel block in the similar area, and obtain the noise correlation of each block area in combination with the possibility of noise pixels appearing in each superpixel block, and obtain the possibility that the block area is a noise area according to the similarity and noise correlation between the block area and the co-located area;
[0010] Obtain a gradient image of the band image, obtain the gradient value of the superpixel block of the gradient image, cluster the superpixel blocks according to the gradient value, and obtain the probability that each superpixel block is a wrinkle superpixel block according to the distance between the superpixel block and the cluster center and the difference in the adjacent gradient values of the superpixel blocks; and obtain the final noise probability of the block area according to the probability that each superpixel block is a wrinkle superpixel block, the possibility that noise pixels appear in the superpixel block, and the possibility that the block area is a noise area;
[0011] De-noising is performed based on the final noise probability and color fastness testing is completed.
[0012] Preferably, the method for obtaining a plurality of pixel points with the largest difference according to the superpixel block and similar superpixel blocks is:
[0013] Sort all the pixels of the superpixel block from large to small according to the pixel value to obtain a pixel value sequence. Sort the similar pixel blocks of the superpixel block from small to large according to the pixel value. For any superpixel block and its similar superpixel block, subtract the pixel values of the two pixels at the same position in the pixel value sequence to obtain a preset number of pixels with the largest pixel value difference.
[0014] Preferably, the method for obtaining the possibility of noise pixels appearing in each super pixel block according to the plurality of pixels with the largest differences and the number of similar super pixel blocks is:
[0015]
[0016] In the formula, represents the relationship between the kth superpixel block and the vth similar pixel block. The absolute value of the grayscale difference of pixels with large grayscale differences. is the number of pixels with large differences, represents the number of similar superpixel blocks to the kth superpixel block, Indicates the possibility of noise pixels appearing in the kth superpixel block.
[0017] Preferably, the method for obtaining the co-located area of the block area is:
[0018] For each band image, the center point of each block area is obtained, and the block area corresponding to the center point is obtained in the remaining band images. The block areas of the center point of the block area of each band image in the remaining band images are recorded as the co-located areas.
[0019] Preferably, the method for obtaining the difference value according to the grayscale value of each super pixel block in the block area is:
[0020] For each block area of the band image, the difference value of each super pixel block in each block area is obtained by subtracting the gray value mean value of each super pixel block from that of the adjacent super pixel block.
[0021] Preferably, the method for obtaining the similarity between the block region and the co-located region according to the difference value and the contour similarity is:
[0022] Calculate the probability that the number of superpixel blocks corresponding to each different difference value in the block area accounts for the number of superpixel blocks in the block area, and calculate the probability that the number of superpixel blocks corresponding to each different difference value in the co-located area accounts for the number of superpixel blocks in the block area; subtract the probabilities corresponding to the same difference value between the block area and the co-located area, add the differences of the probabilities corresponding to all the difference values, and then multiply them by the contour similarity to obtain the similarity between the block area and the co-located area.
[0023] Preferably, the method of obtaining the grayscale area difference according to the grayscale difference between each block area and the super pixel block in the similar area, and obtaining the noise correlation degree of each block area in combination with the possibility of noise pixels appearing in each super pixel block is:
[0024]
[0025] In the formula, is the possibility that there are noise pixels in the u-th superpixel block of the c-th similar area of the z-th block area, represents the number of superpixel blocks in the c-th similar region, It represents the grayscale difference between the zth block area and its cth similar area super pixel block. is the linear normalization function, It represents the noise correlation between the z-th block area and its c-th similar area.
[0026] Preferably, the method for obtaining the probability that each superpixel block is a wrinkle superpixel block according to the distance between the superpixel block and the cluster center and the difference in adjacent gradient values of the superpixel block is:
[0027]
[0028] In the formula, represents the Euclidean distance between the kth superpixel block and the a1th central pixel block, represents the gradient difference between the kth superpixel block and its adjacent a2th superpixel block, represents the number of central pixel blocks, represents the number of neighboring superpixel blocks of the kth superpixel block, represents the linear normalization function, represents the probability that the kth superpixel block is a wrinkled superpixel block.
[0029] The beneficial effect of the present invention is that the existing Gaussian filtering algorithm performs denoising on the global image, but cannot guarantee that each area in the denoised image can maintain a good denoising effect. The present invention combines different band images for analysis, combines similar areas in each band image to obtain the possibility of pixel block noise in each area, and combines the grayscale changes of pixel blocks that may be caused during the color fastness experiment, thereby constructing a regional noise degree index, and performing secondary denoising on the high noise level according to the noise level of each area. Through the denoising process of the present invention, it is ensured that the image is denoised as a whole, and the individual high noise point areas of the image are denoised twice. At the same time, the noise area determination accuracy is improved by combining the surface changes of clothing fabrics before and after the experiment, which greatly improves the image denoising effect and ensures the experimental reliability and accuracy when performing color fastness detection based on spectral images. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0031] Figure 1 A schematic flow chart of a method for testing the performance of clothing fabrics provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0033] See also Figure 1 , which shows a flow chart of a method for testing the performance of clothing fabrics provided by an embodiment of the present invention, the method comprising the following steps:
[0034] Step S001, acquiring a spectral image according to a spectral image acquisition device and a neural network technology.
[0035] The existing spectral image acquisition equipment is used to collect the corresponding images of the clothing fabric. After the clothing fabric is subjected to operations such as sweat immersion, the spectral image of the clothing fabric is collected by the spectral image acquisition equipment. The collected spectral image is semantically segmented using a neural network, and the pixels at the clothing fabric are marked as 1, and the background pixels are marked as 0. Several training images are input, and the loss function is a cross entropy loss function. The output is a segmented spectral image, and the collected spectral image is semantically segmented to obtain information that only contains pixels of clothing fabric in the spectral image.
[0036] At this point, a spectral image containing only clothing fabric information is obtained.
[0037] Step S002, obtaining the possibility of noise in each superpixel block based on the difference between the superpixel block in the same band and the similar superpixel block, and obtaining the possibility that the block area is a noise area based on the difference between the superpixel blocks in the block areas of different bands and the similarity of the block areas.
[0038] For the spectral image, the SLIC superpixel segmentation algorithm is used to segment it to obtain a number of superpixel blocks. The size of the superpixel block in this embodiment is 10*10, and the average pixel value of each superpixel block is obtained. The SLIC superpixel segmentation is a well-known technology and will not be described in detail here.
[0039] For spectral images, the principal component analysis method is used to reduce the dimension of the spectral image to obtain several band images. Each band image is a single-layer image of the spectral image in different bands. At the same time, since clothing fabrics may have patterns, etc., regional analysis of the band images can obtain more accurate color fastness results. First, each superpixel block is regarded as a pixel point to obtain a histogram. The DBSCAN algorithm is used to cluster the pixel values of the superpixel blocks in the histogram to obtain several clusters. For each cluster, the region growing method is used based on any superpixel block, and the area after region growth is used as the block area.
[0040] For any two superpixel blocks, the sum of the pixel values of all pixels in the superpixel blocks and the average pixel value are subtracted to obtain the similarity between the two superpixel blocks. A similarity threshold is set. If the similarity is greater than the similarity threshold, it means that the two superpixel blocks are similar.
[0041] For each superpixel block, sort all the pixels of the superpixel block from large to small according to the pixel value to obtain a pixel value sequence. For the similar pixel blocks of the superpixel block, sort them from small to large according to the pixel value. For any superpixel block and its similar superpixel block, subtract the pixel values of the two pixels at the same position in the pixel value sequence to obtain the pixel with the largest pixel value difference. pixels, in this embodiment, let The value is 5, which is based on the superpixel block and the largest difference among similar superpixel blocks. The probability of noise pixels appearing in each superpixel block is obtained by the number of pixels and similar superpixel blocks. The formula is as follows:
[0042]
[0043] In the formula, represents the relationship between the kth superpixel block and the vth similar pixel block. The absolute value of the grayscale difference of pixels with large grayscale differences. is the number of pixels with large differences, represents the number of similar superpixel blocks to the kth superpixel block, Indicates the possibility of noise pixels appearing in the kth superpixel block.
[0044] The greater the difference between the pixel with the largest grayscale value difference between the super pixel block and its similar super pixel block, the more likely the current pixel block contains noise pixels. The larger the value, the larger the E value, which means that the superpixel block is more likely to have noise. The reason for this is to avoid the possibility that the noise obtained based on other pixels is not obvious, so a small number of pixels with large differences are selected for analysis.
[0045] For the same material area, for different band images, the grayscale values of adjacent pixels will not change significantly. Therefore, the possibility of noise pixels in the spectral image can be obtained based on the different band images.
[0046] For a block area in a certain band image, the center point position of the block area is obtained, and the block area corresponding to the same center point is found in another band image. The area with the same center point is called the co-located area. The shape context algorithm is used to obtain the contour similarity of the two block areas. The shape context operator is a well-known technology and will not be described in detail here. For each block area of the band image, the difference between each super pixel block in each block area and the gray value mean of the adjacent super pixel block is subtracted to obtain the difference value, and the difference value corresponding to each super pixel block is calculated. The probability of the number of super pixel blocks corresponding to each difference value accounting for all super pixel blocks in the block area is calculated. For the co-located area of the block area, the difference value corresponding to each super pixel block is also obtained. The probability corresponding to each difference value of the block area and its co-located area is subtracted and combined with the contour similarity to judge the similarity of the block area and its co-located area. The formula is as follows:
[0047]
[0048] In the formula, Indicates the maximum value of the difference between the superpixel block in the block area and its co-located area. represents the probability corresponding to the i-th difference value of the z-th block area, represents the probability corresponding to the i-th difference value of the z-th block area and its s-th co-located area, represents the contour similarity between the zth block region and its sth co-located region, Indicates the regional similarity between the z-th block region and its s-th co-located region.
[0049] The more similar the contours of the two regions are and the more similar the difference relationship between each superpixel block in the two regions and the surrounding area is, the greater the possibility that the block region is similar to the co-located region. The greater the similarity possibility, the less likely the block region is affected by noise.
[0050] Set similarity threshold here , for each block area of the band image, the regional similarity is obtained and linearly normalized. When the normalized value is greater than the similarity threshold , the block area is considered to be similar to the co-located area, and the noise level of the block area is judged based on the block area and its similar co-located area.
[0051] All similar areas of each block area are counted, and the difference between the grayscale value of each superpixel block in the block area and the average grayscale value of the block area is recorded as the grayscale difference. The grayscale difference obtained by each superpixel block in the block area and the grayscale difference obtained by each superpixel block in the similar area are recorded as the grayscale area difference. According to the grayscale area difference and the possibility that each superpixel block in the similar area has a noise pixel point, the noise correlation of the block area is obtained. The formula is as follows:
[0052]
[0053] In the formula, is the possibility that there are noise pixels in the u-th superpixel block of the c-th similar area of the z-th block area, represents the number of superpixel blocks in the c-th similar region, It represents the grayscale difference between the zth block area and its cth similar area super pixel block. is the linear normalization function, It represents the noise correlation between the z-th block area and its c-th similar area.
[0054] The greater the possibility that each superpixel block has noise pixels, the greater the noise correlation between its block area and its similar area. It can be regarded as a weight, and its value represents the difference in grayscale value between the block area and the similar area. The smaller the difference, the more related the block image is to the superpixel block in the similar area, that is, The larger the value is, the greater the noise correlation between the z-th block area and its similar areas.
[0055] According to all similar regions and regional similarities of the block area, the possibility that the block area is a noise area is obtained. The formula is as follows:
[0056]
[0057] In the formula, represents the noise correlation between the zth block area and its cth similar area, represents the regional similarity between the zth block region and its cth similar region, represents the linear normalization function, Indicates the number of similar regions owned by the z-th block region, Indicates the possibility that the zth block area is a noise area.
[0058] Represents the weight. The greater the regional similarity, the more influence it represents. The greater the noise correlation between the block area and the similar area, the greater the possibility that the block area is a noise area.
[0059] Step S003, obtaining the probability that each superpixel block is a wrinkle superpixel block according to the gradient value of the gradient image of the band image, and based on this, obtaining the final noise possibility of the block area, and completing denoising based on the final noise possibility.
[0060] When testing color fastness, factors such as friction and sweat stains may cause wrinkles and creases on the clothing fabric. In the process of noise recognition, these pixels may be mistakenly identified as noise pixels, resulting in inaccurate final results. Therefore, it is necessary to identify the possible wrinkle areas based on these situations.
[0061] The texture analysis of the band image is performed, that is, the gradient image is analyzed, and the gradient image of the band image is obtained using the Sobel operator. If the clothing surface has the above-mentioned wrinkles and other factors, then the pixel blocks with large gradient changes in the corresponding area should be more concentrated, based on which the probability of each superpixel block being a wrinkle superpixel block can be calculated.
[0062] Specifically, for the gradient image obtained from the band image, the superpixel blocks in the gradient image are clustered using the DBSCAN algorithm, where the clustering distance is the gradient difference of the superpixel blocks in the gradient image. The Sobel algorithm and the DBSCAN algorithm for obtaining the gradient image are both well-known algorithms and will not be described in detail here. Set the gradient threshold , in this embodiment, the gradient threshold , after clustering, the mean of the gradient value of each cluster is calculated. If the mean is greater than the gradient threshold, the cluster center of the cluster is obtained, and the cluster center is recorded as the central pixel block. For each superpixel block, the distance between the superpixel block and the cluster center is calculated. If the distance is closer, the superpixel blocks with larger gradient changes are more concentrated, and the probability that the superpixel block is a wrinkled superpixel block is greater. In addition, the gradient value difference between the superpixel block and the surrounding superpixel blocks is obtained. If the gradient value difference is smaller, that is, the gradient value of the surrounding superpixel blocks is closer to that of the superpixel block, it means that the possibility that the superpixel block is a wrinkled superpixel block is greater. Therefore, the probability that each superpixel block is a wrinkled superpixel block is obtained, and the formula is as follows:
[0063]
[0064] In the formula, represents the Euclidean distance between the kth superpixel block and the a1th central pixel block, represents the gradient difference between the kth superpixel block and its adjacent a2th superpixel block, represents the number of central pixel blocks, represents the number of neighboring superpixel blocks of the kth superpixel block, represents the linear normalization function, represents the probability that the kth superpixel block is a wrinkled superpixel block.
[0065] From this, the final noise probability of each block area can be calculated as follows:
[0066]
[0067] In the formula, It indicates the possibility of noise pixels appearing in the u-th superpixel block in the z-th block area. represents the probability of the u-th superpixel block in the z-th block area being a wrinkled superpixel block, is the number of superpixel blocks in the z-th partition area, Indicates the possibility that the z-th block area is a noise area, is the linear normalization function, represents the final noise likelihood for the zth region.
[0068] Set the denoising threshold , in this embodiment, ,If the final noise probability of each block area is greater than the denoising threshold, the block is denoised by Gaussian filtering and smoothing, and each band image is denoised separately.
[0069] At this point, the noise removal of each band image is completed.
[0070] Step S004, after denoising, calculate the color fastness value of each band image pixel to complete the color fastness detection of the spectral image.
[0071] After the clothing color fastness experiment, spectral images are collected again, and the spectral images are denoised using the above method. For the spectral images before and after the color fastness experiment, the grayscale difference of the pixel is calculated once for each band image in each band. The color fastness value of each pixel is obtained by the ratio of the absolute value of the grayscale difference of the pixel in each band image to the probability that the pixel is in the wrinkle superpixel block. The color fastness value of each pixel in the band images corresponding to all bands in each spectral image is linearly normalized, and the normalized values of all pixels are averaged to obtain the color fastness value of the spectral image.
[0072] Setting standard thresholds , in this embodiment, If the color fastness value in the spectral image is greater than the standard threshold, it is considered that the color fastness of the clothing fabric does not meet the standard, otherwise it meets the standard. At this point, the color fastness test of the clothing fabric is completed.
[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for testing the performance of clothing fabrics, characterized in that: The method comprises the following steps: Acquire spectral images; The spectral image is segmented by superpixel to obtain several superpixel blocks, several band images of the spectral image are obtained, the band images are clustered to obtain block areas for the superpixel blocks, similar superpixel blocks of the superpixel blocks are obtained, several pixels with the largest differences are obtained according to the superpixel blocks and similar superpixel blocks, and the possibility of noise pixels appearing in each superpixel block is obtained according to the several pixels with the largest differences and the number of similar superpixel blocks; Obtain the co-located area of the block area, obtain the contour similarity of the block area and the co-located area, obtain the difference value according to the grayscale value of each superpixel block in the block area, obtain the similarity of the block area and the co-located area according to the difference value and the contour similarity, obtain the grayscale area difference according to the grayscale difference between each block area and the superpixel block in the similar area, and obtain the noise correlation of each block area in combination with the possibility of noise pixels appearing in each superpixel block, and obtain the possibility that the block area is a noise area according to the similarity and noise correlation between the block area and the co-located area; Obtain a gradient image of the band image, obtain the gradient value of the superpixel block of the gradient image, cluster the superpixel blocks according to the gradient value, and obtain the probability that each superpixel block is a wrinkle superpixel block according to the distance between the superpixel block and the cluster center and the difference in the adjacent gradient values of the superpixel blocks; and obtain the final noise probability of the block area according to the probability that each superpixel block is a wrinkle superpixel block, the possibility that noise pixels appear in the superpixel block, and the possibility that the block area is a noise area; De-noising is performed based on the final noise probability and color fastness testing is completed; The expression of the final noise probability of the block area is: In the formula, It indicates the possibility of noise pixels appearing in the u-th superpixel block in the z-th block area. represents the probability that the u-th superpixel block in the z-th partition area is a wrinkled superpixel block, is the number of superpixel blocks in the z-th partition area, Indicates the possibility that the z-th block area is a noise area, is the linear normalization function, represents the final noise likelihood for the zth region.
2. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining a plurality of pixel points with the largest difference based on a super pixel block and similar super pixel blocks is: Sort all the pixels of the superpixel block from large to small according to the pixel value to obtain a pixel value sequence. Sort the similar pixel blocks of the superpixel block from small to large according to the pixel value. For any superpixel block and its similar superpixel block, subtract the pixel values of the two pixels at the same position in the pixel value sequence to obtain a preset number of pixels with the largest pixel value difference.
3. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining the possibility of noise pixels appearing in each super pixel block according to the number of pixels with the largest differences and the number of similar super pixel blocks is: In the formula, represents the relationship between the kth superpixel block and the vth similar pixel block. The absolute value of the grayscale difference of pixels with large grayscale differences. is the number of pixels with large differences, represents the number of similar superpixel blocks to the kth superpixel block, Indicates the possibility of noise pixels appearing in the kth superpixel block.
4. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining the co-located area of the block area is: For each band image, the center point of each block area is obtained, and the block area corresponding to the center point is obtained in the remaining band images. The block areas of the center point of the block area of each band image in the remaining band images are recorded as the co-located areas.
5. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining the difference value according to the grayscale value of each super pixel block in the block area is: For each block area of the band image, the difference value of each super pixel block in each block area is obtained by subtracting the gray value mean value of each super pixel block from that of the adjacent super pixel block.
6. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining the similarity between the block region and the co-located region according to the difference value and the contour similarity is: Calculate the probability that the number of superpixel blocks corresponding to each different difference value in the block area accounts for the number of superpixel blocks in the block area, and calculate the probability that the number of superpixel blocks corresponding to each different difference value in the co-located area accounts for the number of superpixel blocks in the block area; subtract the probabilities corresponding to the same difference value between the block area and the co-located area, add the differences of the probabilities corresponding to all the difference values, and then multiply them by the contour similarity to obtain the similarity between the block area and the co-located area.
7. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method of obtaining the grayscale area difference according to the grayscale difference between each block area and the super pixel block in the similar area, and obtaining the noise correlation degree of each block area in combination with the possibility of noise pixels appearing in each super pixel block is as follows: In the formula, is the possibility that there are noise pixels in the u-th superpixel block of the c-th similar area of the z-th block area, represents the number of superpixel blocks in the c-th similar region, It represents the grayscale difference between the zth block area and its cth similar area super pixel block. is the linear normalization function, It represents the noise correlation between the z-th block area and its c-th similar area.
8. A method for testing the performance of clothing fabrics according to claim 1, characterized in that: The method for obtaining the probability that each superpixel block is a wrinkle superpixel block according to the distance between the superpixel block and the cluster center and the difference in adjacent gradient values of the superpixel block is: In the formula, represents the Euclidean distance between the kth superpixel block and the a1th central pixel block, represents the gradient difference between the kth superpixel block and its adjacent a2th superpixel block, represents the number of central pixel blocks, represents the number of neighboring superpixel blocks of the kth superpixel block, represents the linear normalization function, represents the probability that the kth superpixel block is a wrinkled superpixel block.
