A fabric structure image recognition method

By combining Markov random field models and convolutional neural networks, the problem of low accuracy in fabric structure image recognition was solved, achieving efficient and stable recognition in complex environments.

CN118864885BActive Publication Date: 2024-11-29HAIMEN QIWEI SWEATER WEAVING
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411346230.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-26
Publication Date
2024-11-29
Estimated Expiration
2044-09-26

AI Technical Summary

Technical Problem

In existing technologies, image recognition of fabric structures suffers from low accuracy, especially in complex environments and under different lighting conditions where texture features are unstable, and imbalanced samples lead to poor model recognition performance.

Method used

The MRF fabric structure texture image is obtained by using a Markov random field model to obtain the grayscale image of the fabric surface. The fabric structure is identified by calculating the local texture feature difference coefficient and the local structural complexity coefficient, combined with a convolutional neural network model.

Benefits of technology

It improves the accuracy of fabric structure image recognition by refining texture features and considering the complex structural features of yarn interweaving in local areas, thereby enhancing the stability and accuracy of recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118864885B_ABST
    Figure CN118864885B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of image recognition, and proposes a method for fabric structure image recognition, including: obtaining a fabric surface grayscale image; obtaining an MRF fabric structure texture image based on the fabric surface grayscale image, and calculating the fabric yarn local structure complexity coefficient of each pixel point based on the local structure difference characteristics in the MRF fabric structure texture image; obtaining a fabric local structure feature image based on the fabric yarn local structure complexity coefficient of each pixel point, and obtaining a fabric local structure tightness coefficient based on the fabric local structure feature image; obtaining a fabric structure discriminant vector based on the fabric local structure tightness coefficient, and obtaining a fabric structure image recognition result based on the fabric structure discriminant vector. The present application improves the accuracy of image recognition of fabric structure by constructing a fabric texture uniformity discriminant coefficient.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of image recognition, and in particular to a method for recognizing a fabric structure image. Background Art

[0002] In the textile manufacturing industry, fabric production has shown a trend of small batches and multiple varieties. Textile factories need a fast and accurate method for fabric structure identification. The existing analysis and detection of fabric structure not only consumes manpower and material resources, but also has low work efficiency. In addition, the test results obtained in complex testing environments may vary greatly. With the rapid development of automation technology, the use of computer vision technology to identify fabric structure can improve work efficiency and detection accuracy, and provide objective and accurate fabric structure information for the next step of fabric dyeing.

[0003] However, there are some defects in using image recognition to identify fabric structure, mainly including: different types of fabrics have complex and diverse textures and structures, some fabrics even have large differences at the microscopic level, and the appearance of the fabric will be affected by lighting and shooting angle. The same piece of fabric may show different texture characteristics under different lighting conditions, which will lead to instability in image recognition. At the same time, in practical applications, the number of samples of different types of fabrics may be unbalanced, which will cause the trained model to perform poorly when identifying minority categories. Appropriate methods need to be taken to solve the problem of sample imbalance. Therefore, due to environmental characteristics and structural characteristics of the fabric, the accuracy of image recognition of fabric structure is low. Summary of the invention

[0004] The present application provides a method for recognizing a fabric structure image to solve the problem of low accuracy in recognizing a fabric structure image. The technical solution adopted is as follows:

[0005] An embodiment of the present application provides a method for recognizing a fabric structure image, the method comprising the following steps:

[0006] Acquire a fabric surface grayscale image, and acquire an MRF fabric structure texture image according to the fabric surface grayscale image;

[0007] According to the data randomness and distribution difference of the local area where each pixel point is located in the MRF fabric structure texture image, the fabric local texture feature vector and the local feature probability distribution curve of each pixel point are obtained; according to the fabric local texture feature vector and the local feature probability distribution curve of each pixel point in the MRF fabric structure texture image, the fabric local texture feature difference coefficient of each pixel point is calculated;

[0008] Calculate the fabric yarn local structure complexity coefficient of each pixel point according to the fabric local texture feature difference coefficient of each pixel point in the MRF fabric structure texture image; obtain the fabric local structure analysis area according to the fabric yarn local structure complexity coefficient of each pixel point in the MRF fabric structure texture image; calculate the fabric local structure compactness coefficient of each fabric local structure analysis area according to the difference of the fabric yarn local structure complexity coefficient of the pixel points between the fabric local structure analysis areas;

[0009] A fabric structure discriminant vector is obtained according to the fabric local structure compactness coefficient of each fabric local structure analysis area; and a fabric structure image recognition result is obtained according to the fabric structure discriminant vector.

[0010] Preferably, the method of obtaining the fabric surface grayscale image and obtaining the MRF fabric structure texture image according to the fabric surface grayscale image is:

[0011] Using a camera to photograph the textile surface, obtaining an RGB image of the textile surface, converting the acquired RGB image of the textile surface into a grayscale image, and using the converted grayscale image as the textile surface grayscale image; using an image denoising algorithm and an image enhancement algorithm to process the textile surface grayscale image respectively, and using the processed result image as the fabric surface grayscale image;

[0012] A texture synthesis image of the fabric surface grayscale image is obtained by using a Markov random field model, and the texture synthesis image is used as an MRF fabric structure texture image.

[0013] Preferably, the method for obtaining the fabric local texture feature vector and local feature probability distribution curve of each pixel point according to the data randomness and distribution difference of the local area where each pixel point is located in the MRF fabric structure texture image is:

[0014] A local feature window of preset size is constructed with each pixel in the MRF fabric structure texture image as the center, and the pixels within the 8-neighborhood of each pixel in the MRF fabric structure texture image are taken as the neighboring feature pixels of each pixel;

[0015] Obtain the fabric local texture feature vector of each pixel point according to the data randomness feature of each row of pixels in the local feature window of each pixel point in the MRF fabric structure texture image;

[0016] The probability distribution statistics of the grayscale values ​​of all pixels in the local feature window of each pixel in the MRF fabric structure texture image are performed to obtain the local feature probability distribution curve corresponding to each pixel in the MRF fabric structure texture image.

[0017] Preferably, the method for obtaining the fabric local texture feature vector of each pixel point according to the data randomness feature of each row of pixel points in the local feature window of each pixel point in the MRF fabric structure texture image is:

[0018] The grayscale value of each row of pixels in the local feature window of each pixel is taken as input, and the permutation entropy algorithm is used to obtain the permutation entropy of the grayscale value of each row of pixels in the local feature window of each pixel; the vector composed of the permutation entropies corresponding to all rows of pixels in the local feature window of each pixel in ascending order is taken as the local texture feature vector of the fabric.

[0019] Preferably, the method for calculating the fabric local texture feature difference coefficient of each pixel point according to the fabric local texture feature vector and the local feature probability distribution curve of each pixel point in the MRF fabric structure texture image is:

[0020]

[0021] In the formula, Indicates The fabric local texture feature difference coefficient of each pixel; Indicates The local texture feature vector of fabric with pixels, Indicates The pixel The local texture feature vector of the fabric with neighboring feature pixels, express and The cosine similarity between ; Indicates The local feature probability distribution curve of pixels, Indicates The pixel The local feature probability distribution curve of neighboring feature pixels, express and The KL divergence between Indicates The number of neighboring feature pixels of a pixel.

[0022] Preferably, the method for calculating the local structure complexity coefficient of the fabric yarn at each pixel point according to the fabric local texture feature difference coefficient of each pixel point in the MRF fabric structure texture image is:

[0023]

[0024] In the formula, Indicates The local structural complexity coefficient of the fabric yarn at each pixel point; Indicates The fabric local texture feature difference coefficient of pixels, Indicates The pixel The fabric local texture feature difference coefficient of neighboring feature pixels; Represents the first pixels, Indicates The local feature window of pixels pixels, express and The Euclidean distance between Indicates The mean value of the fabric local texture feature difference coefficients of all pixels in the local feature window of pixels; Indicates The number of pixels in the local feature window of pixels.

[0025] Preferably, the method for obtaining the fabric local structure analysis area according to the fabric yarn local structure complexity coefficient of each pixel point in the MRF fabric structure texture image is:

[0026] The grayscale value of each pixel in the MRF fabric structure texture image is updated by using the local structure complexity coefficient of the fabric yarn. The result image after the grayscale value in the MRF fabric structure texture image is updated is used as the fabric local structure feature image and input as the fabric local structure feature image. The watershed algorithm is used to obtain the segmentation result image of the fabric local structure feature image, and each image block in the segmentation result of the fabric local structure feature image is used as a fabric local structure analysis area.

[0027] Preferably, the method for calculating the fabric local structure compactness coefficient of each fabric local structure analysis area according to the difference in fabric yarn local structure complexity coefficients of pixel points between fabric local structure analysis areas is:

[0028]

[0029] In the formula, Indicates The fabric local structure compactness coefficient of the fabric local structure analysis area; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent and The grayscale run-length matrix of the local structure analysis area of ​​the fabric, express and SAD value between ; Indicates the number of fabric local structure analysis areas.

[0030] Preferably, the method for obtaining the fabric structure discriminant vector according to the fabric local structure compactness coefficient of each fabric local structure analysis area is:

[0031] A set of fabric local structure tightness coefficients of all fabric local structure analysis areas of the fabric local structure feature image is taken as a fabric structure discrimination data set and input as a fabric structure discrimination data set. A normalization algorithm is used to obtain the normalized result of each element in the fabric structure discrimination data set. A vector composed of the normalized results of all elements in the fabric structure discrimination data set in ascending order is taken as a fabric structure discrimination vector.

[0032] Preferably, the method for obtaining the fabric structure image recognition result according to the fabric structure discriminant vector is:

[0033] The input is the acquired grayscale image of the fabric surface and the corresponding fabric structure discriminant vector. The convolutional neural network model is used to obtain the judgment result of the fabric structure. The judgment result is the probability that the acquired grayscale image of the fabric surface is plain, twill and satin. The recognition of the fabric structure image is completed according to the judgment result of the fabric structure.

[0034] The beneficial effects of the present application are: using a Markov random field model to obtain an MRF fabric structure texture image of a fabric surface grayscale image, refining the texture features of a plant tissue structure through the MRF fabric structure texture image, and constructing a fabric yarn local structure complexity coefficient based on the texture feature differences of local areas in the MRF fabric structure texture image, reflecting the complex structural features of yarn interlacing in a local area of ​​the fabric tissue structure through the fabric yarn local structure complexity coefficient, and constructing a fabric local structure feature image through the fabric yarn local structure complexity coefficient, and obtaining a fabric local structure compactness coefficient based on the fabric local structure feature image, reflecting the dense features of local interlacing of the fabric tissue structure through the fabric local structure compactness coefficient, and the beneficial effects are that the distribution of complex structural features of local yarn interlacing is reflected through the fabric local structure feature image, and the accuracy of fabric tissue structure image recognition is improved by considering the distribution of complex structural features of yarn interlacing and combining the convolutional neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A schematic flow chart of a method for identifying a fabric structure image provided by an embodiment of the present application;

[0037] Figure 2 A schematic diagram of the implementation process of obtaining fabric structure image recognition results provided in one embodiment of the present application. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0039] See also Figure 1 , which shows a flowchart of a method for recognizing a fabric structure image provided by an embodiment of the present application, the method comprising the following steps:

[0040] Step S001, obtaining a grayscale image of the fabric surface.

[0041] In the process of automated production and inspection of textiles, a CCD camera is used to photograph the surface of the textile to obtain an RGB image of the textile surface, and the collected RGB image of the textile surface is converted into a grayscale image. The specific implementation process of converting the RGB image into the grayscale image is a well-known technology and will not be described in detail. The converted grayscale image is used as the grayscale image of the textile surface. Since the grayscale image of the textile surface may be greatly affected by noise due to the influence of ambient light and camera parameters during the shooting process, which affects the analysis of the surface structural characteristics of the textile, the input is the grayscale image of the textile surface, and a median filtering algorithm is used to obtain a denoised result image of the grayscale image of the textile surface. The denoised result image of the grayscale image of the textile surface is used as the input of an adaptive histogram equalization algorithm to obtain an image enhanced result image of the grayscale image of the textile surface, and the image enhanced result image of the grayscale image of the textile surface is used as the grayscale image of the fabric surface. Both the median filtering algorithm and the adaptive histogram equalization algorithm are well-known technologies and the specific process will not be described in detail.

[0042] At this point, the grayscale image of the fabric surface is obtained.

[0043] Step S002, obtaining an MRF fabric structure texture image according to the fabric surface grayscale image, and calculating the fabric yarn local structure complexity coefficient of each pixel point according to the local structure difference characteristics in the MRF fabric structure texture image.

[0044] The fabric structure is generally divided into three types: plain, twill and satin, and the connection density based on the direction of the three structures is quite different. Specifically, the satin structure is the tightest, while the twill and plain structures are relatively loose. Therefore, the tightness of the fabric structure is calculated based on the difference characteristics of the density of the warp and weft connection. Since the distribution of yarns in the local structure of the fabric is only affected by the distribution of yarns in adjacent positions, that is, if the yarn connection is too loose or too tight, the adjacent yarns will be squeezed or pulled, affecting the fabric structure characteristics.

[0045] Furthermore, since the Markov random field can capture the high-order statistical characteristics of the original texture when performing texture synthesis and retains the important visual texture information in the original texture features, the Markov random field model (Markov Random Field, MRF) is used to obtain the texture synthesis image of the grayscale image of the fabric surface, and the texture synthesis image is used as the MRF fabric structure texture image. The Markov random field model is a well-known technology, and the specific implementation process will not be repeated here.

[0046] Furthermore, each pixel in the MRF fabric structure texture image is used as the center to construct The local feature window of size is an odd number and its size is an empirical value of 11. The grayscale value of each row of pixels in the local feature window of each pixel is taken as input, and the permutation entropy algorithm is used to obtain the permutation entropy of the grayscale value of each row of pixels in the local feature window of each pixel. The permutation entropy algorithm reflects the randomness of the grayscale value of each row of pixels in the local feature window of each pixel. The permutation entropy algorithm is a well-known technology, and the specific calculation process is not repeated here. The permutation entropy corresponding to all rows of pixels in the local feature window of each pixel is used as a vector composed of the order from small to large as the local texture feature vector of the fabric; the grayscale value in the local feature window of each pixel is subjected to probability statistics to obtain the local feature probability distribution curve corresponding to each pixel; the pixels in the neighborhood of each pixel are taken as the neighboring feature pixels of each pixel, and the fabric local texture feature difference coefficient is calculated according to the difference of the fabric local texture features between each pixel and the corresponding neighboring feature pixels. The specific calculation formula is as follows:

[0047]

[0048] In the formula, Indicates The fabric local texture feature difference coefficient of each pixel; Indicates The local texture feature vector of fabric with pixels, Indicates The pixel The local texture feature vector of the fabric with neighboring feature pixels, express and The cosine similarity between ; Indicates The local feature probability distribution curve of pixels, Indicates The pixel The local feature probability distribution curve of neighboring feature pixels, express and The KL divergence between Indicates The number of neighboring feature pixels of a pixel.

[0049] Jordi The local texture features of the pixel point and its neighboring feature pixel points are quite different, and the gray value distribution of the pixel points is quite different, then the calculated and The larger the value of The difference coefficient of fabric local texture feature at pixel points The larger the value, the more Compare the grayscale distribution and random variation characteristics of the pixel points in the local area with those of the neighboring feature pixels. The more significant the difference in local texture features of fabric in the local area of ​​​​the pixel points, the more significant the difference.

[0050] Furthermore, the local structural complexity coefficient of the fabric yarn is calculated according to the difference between the pixel point in the local feature window of each pixel point and the central pixel point in the local feature window. The local structural complexity coefficient of the fabric yarn reflects the complexity of the fabric structure texture in the local area of ​​each pixel point. The specific calculation formula is as follows:

[0051]

[0052] In the formula, Indicates The local structural complexity coefficient of the fabric yarn at each pixel point; Indicates The fabric local texture feature difference coefficient of pixels, Indicates The pixel The fabric local texture feature difference coefficient of neighboring feature pixels; Represents the first pixels, Indicates The local feature window of pixels pixels, express and The Euclidean distance between Indicates The mean value of the fabric local texture feature difference coefficients of all pixels in the local feature window of pixels; Indicates The number of pixels in the local feature window of pixels.

[0053] Jordi The fabric local texture feature difference coefficients between the pixels and the corresponding pixels in the local feature window are quite different, and the fabric local texture feature difference between the pixels with closer Euclidean distance is greater, that is, the calculated The larger the value of The smaller the value of The local structural complexity coefficient of the fabric yarn at pixel points The larger the value is, the stronger the MRF fabric structure texture image is. The more complex the fabric texture structure is in the local area of ​​the pixel, the more obvious it is.

[0054] So far, the local structural complexity coefficient of the fabric yarn at each pixel point in the MRF fabric structure texture image is obtained.

[0055] Step S003, obtaining a fabric local structure feature image according to the fabric yarn local structure complexity coefficient of each pixel point, and obtaining a fabric local structure compactness coefficient according to the fabric local structure feature image.

[0056] Plain, twill and satin fabrics are all composed of different primitive structures, which also reflect the distribution characteristics of fabric texture. The fabric structure can be judged according to the distribution characteristics of different fabric textures. The primitive structure of the fabric appears repeatedly on the fabric surface, so the fabric texture distribution information can be extracted by the change law of the local structure complexity coefficient of the fabric yarn of the pixel point in the MRF fabric structure texture image. If it is a plain structure, the local connection complexity characteristics of the fabric structure are less significant, and the connection tightness is poor. At the same time, the tightness and local complex structure of plain, twill and satin gradually increase.

[0057] Furthermore, the grayscale value of each pixel in the MRF fabric structure texture image is updated using the local structure complexity coefficient of the fabric yarn, and the result image after the grayscale value in the MRF fabric structure texture image is updated is used as the fabric local structure feature image. The fabric local structure feature image is used to reflect the complex structural feature differences of plain, twill or satin that may appear in the fabric. The input is the fabric local structure feature image, and the watershed algorithm is used to obtain the segmentation result image of the fabric local structure feature image. Each image block in the segmentation result of the fabric local structure feature image is used as a fabric local structure analysis area. The watershed algorithm is a well-known technology, and the specific calculation process will not be repeated here.

[0058] Further, the local structure complexity coefficient of the fabric yarn of all pixel points in each fabric local structure analysis area is used as input, and the grayscale run matrix algorithm is used to obtain the grayscale run matrix of each fabric local structure analysis area, and the short run advantage value and the long run advantage value of the grayscale run matrix are calculated, wherein the specific calculation process of the grayscale run matrix, the short run advantage value and the long run advantage value is a known technology and will not be repeated. The fabric local structure tightness coefficient of each fabric local structure analysis area is calculated according to the statistical characteristic difference of the smooth change of the fabric yarn local structure complexity coefficient of different fabric local structure analysis areas, and the specific calculation formula is as follows:

[0059]

[0060] In the formula, Indicates The fabric local structure compactness coefficient of the fabric local structure analysis area; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent and The grayscale run-length matrix of the local structure analysis area of ​​the fabric, express and The SAD (Structural Similarity Index) value between them, the specific calculation process of the SAD value is a well-known technology and will not be repeated here; Indicates the number of fabric local structure analysis areas.

[0061] Jordi The less tightly the complex characteristics of the local texture structure between the local structure analysis area of ​​the fabric and other local structure analysis areas, the better the calculated and The smaller the value of is, and the greater the statistical difference of the complex characteristics of the local structure of the fabric yarn between the local structure analysis areas of the fabric is, the calculated The larger the value of The smaller the value of the fabric local structure tightness coefficient of the fabric local structure analysis area, the smaller the value of the fabric local structure tightness coefficient of the fabric local structure analysis area. The more obvious the compact features of the local structural analysis area of ​​the fabric are, the more likely it is a plain weave feature.

[0062] So far, the local structural tightness coefficient of the fabric has been obtained.

[0063] Step S004, obtaining a fabric structure discriminant vector according to the fabric local structure tightness coefficient, and obtaining a fabric structure image recognition result according to the fabric structure discriminant vector.

[0064] A set of fabric local structure tightness coefficients of all fabric local structure analysis areas of the fabric local structure feature image is taken as a fabric structure discrimination data set, and input as a fabric structure discrimination data set. A Z-score normalization algorithm is used to obtain the normalized result of each element in the fabric structure discrimination data set. The Z-score normalization algorithm is a well-known technology, and the specific calculation process will not be repeated here. The normalized results of all elements in the fabric structure discrimination data set are formed in a vector in ascending order as a fabric structure discrimination vector.

[0065] The input is the obtained grayscale image of the fabric surface and the corresponding fabric structure discriminant vector, and the convolutional neural network model is used to obtain the judgment result of the fabric structure, wherein the judgment result is the probability that the obtained grayscale image of the fabric surface is plain, twill and satin, and the recognition of the fabric structure image is completed according to the judgment result of the fabric structure; wherein the loss function of the convolutional neural network model is the cross entropy loss function, the optimization algorithm is the Adam (Adaptivemomentum) optimization algorithm, the convolutional neural network model is a well-known technology, and the specific training process will not be repeated here, and the specific implementation process of obtaining the fabric structure image recognition result is as follows Figure 2 shown.

[0066] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for recognizing a fabric structure image, characterized in that: The method comprises the following steps: Acquire a fabric surface grayscale image, and acquire an MRF fabric structure texture image according to the fabric surface grayscale image; According to the data randomness and distribution difference of the local area where each pixel point is located in the MRF fabric structure texture image, the fabric local texture feature vector and the local feature probability distribution curve of each pixel point are obtained; according to the fabric local texture feature vector and the local feature probability distribution curve of each pixel point in the MRF fabric structure texture image, the fabric local texture feature difference coefficient of each pixel point is calculated; Calculate the fabric yarn local structure complexity coefficient of each pixel point according to the fabric local texture feature difference coefficient of each pixel point in the MRF fabric structure texture image; obtain the fabric local structure analysis area according to the fabric yarn local structure complexity coefficient of each pixel point in the MRF fabric structure texture image; calculate the fabric local structure compactness coefficient of each fabric local structure analysis area according to the difference of the fabric yarn local structure complexity coefficient of the pixel points between the fabric local structure analysis areas; Obtain a fabric structure discriminant vector according to the fabric local structure tightness coefficient of each fabric local structure analysis area; obtain a fabric structure image recognition result according to the fabric structure discriminant vector; The method for calculating the fabric local texture feature difference coefficient of each pixel point according to the fabric local texture feature vector and the local feature probability distribution curve of each pixel point in the MRF fabric structure texture image is: In the formula, Indicates The fabric local texture feature difference coefficient of each pixel; Indicates The local texture feature vector of fabric with pixels, Indicates The pixel The local texture feature vector of the fabric with neighboring feature pixels, express and The cosine similarity between ; Indicates The local feature probability distribution curve of pixels, Indicates The pixel The local feature probability distribution curve of neighboring feature pixels, express and The KL divergence between Indicates The number of neighboring feature pixels of a pixel; The method for calculating the local structure complexity coefficient of the fabric yarn at each pixel point according to the fabric local texture feature difference coefficient of each pixel point in the MRF fabric structure texture image is: In the formula, Indicates The local structural complexity coefficient of the fabric yarn at each pixel point; Indicates The fabric local texture feature difference coefficient of pixels, Indicates The pixel The fabric local texture feature difference coefficient of neighboring feature pixels; Represents the first pixels, Indicates The local feature window of pixels pixels, express and The Euclidean distance between Indicates The mean value of the fabric local texture feature difference coefficients of all pixels in the local feature window of pixels; Indicates The number of pixels in the local feature window of pixels; The method for calculating the fabric local structure compactness coefficient of each fabric local structure analysis area according to the difference in fabric yarn local structure complexity coefficients of pixel points between fabric local structure analysis areas is: In the formula, Indicates The fabric local structure compactness coefficient of the fabric local structure analysis area; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent The short-run advantage value and long-run advantage value of the local structure analysis area of ​​the fabric; and Respectively represent and The grayscale run-length matrix of the local structure analysis area of ​​the fabric, express and SAD value between Indicates the number of fabric local structure analysis areas.

2. A fabric structure image recognition method according to claim 1, characterized in that: The method of obtaining the fabric surface grayscale image and obtaining the MRF fabric structure texture image according to the fabric surface grayscale image is: Using a camera to photograph the textile surface, obtaining an RGB image of the textile surface, converting the acquired RGB image of the textile surface into a grayscale image, and using the converted grayscale image as the textile surface grayscale image; using an image denoising algorithm and an image enhancement algorithm to process the textile surface grayscale image respectively, and using the processed result image as the fabric surface grayscale image; A texture synthesis image of the fabric surface grayscale image is obtained by using a Markov random field model, and the texture synthesis image is used as an MRF fabric structure texture image.

3. A fabric structure image recognition method according to claim 1, characterized in that: The method for obtaining the fabric local texture feature vector and local feature probability distribution curve of each pixel point according to the data randomness and distribution difference of the local area where each pixel point is located in the MRF fabric structure texture image is: A local feature window of preset size is constructed with each pixel in the MRF fabric structure texture image as the center, and the pixels within the 8-neighborhood of each pixel in the MRF fabric structure texture image are taken as the neighboring feature pixels of each pixel; Obtain the fabric local texture feature vector of each pixel point according to the data randomness feature of each row of pixels in the local feature window of each pixel point in the MRF fabric structure texture image; The probability distribution statistics of the grayscale values ​​of all pixels in the local feature window of each pixel in the MRF fabric structure texture image are performed to obtain the local feature probability distribution curve corresponding to each pixel in the MRF fabric structure texture image.

4. A fabric structure image recognition method according to claim 3, characterized in that: The method for obtaining the fabric local texture feature vector of each pixel point according to the data randomness feature of each row of pixels in the local feature window of each pixel point in the MRF fabric structure texture image is: The grayscale value of each row of pixels in the local feature window of each pixel is taken as input, and the permutation entropy algorithm is used to obtain the permutation entropy of the grayscale value of each row of pixels in the local feature window of each pixel; the vector composed of the permutation entropies corresponding to all rows of pixels in the local feature window of each pixel in ascending order is taken as the local texture feature vector of the fabric.

5. The method for recognizing a fabric structure image according to claim 1, characterized in that: The method for obtaining the fabric local structure analysis area according to the fabric yarn local structure complexity coefficient of each pixel point in the MRF fabric structure texture image is: The grayscale value of each pixel in the MRF fabric structure texture image is updated by using the local structure complexity coefficient of the fabric yarn. The result image after the grayscale value in the MRF fabric structure texture image is updated is used as the fabric local structure feature image and input as the fabric local structure feature image. The watershed algorithm is used to obtain the segmentation result image of the fabric local structure feature image, and each image block in the segmentation result of the fabric local structure feature image is used as a fabric local structure analysis area.

6. A fabric structure image recognition method according to claim 1, characterized in that: The method for obtaining the fabric structure discrimination vector according to the fabric local structure tightness coefficient of each fabric local structure analysis area is: A set of fabric local structure tightness coefficients of all fabric local structure analysis areas of the fabric local structure feature image is taken as a fabric structure discrimination data set and input as a fabric structure discrimination data set. A normalization algorithm is used to obtain the normalized result of each element in the fabric structure discrimination data set. A vector composed of the normalized results of all elements in the fabric structure discrimination data set in ascending order is taken as a fabric structure discrimination vector.

7. A fabric structure image recognition method according to claim 1, characterized in that: The method for obtaining the fabric structure image recognition result according to the fabric structure discriminant vector is: The input is the acquired grayscale image of the fabric surface and the corresponding fabric structure discriminant vector. The convolutional neural network model is used to obtain the judgment result of the fabric structure. The judgment result is the probability that the acquired grayscale image of the fabric surface is plain, twill and satin. The recognition of the fabric structure image is completed according to the judgment result of the fabric structure.

Citation Information

Patent Citations

  • Sobel operator-based extraction method of profile and detail composite characteristic vector used for representing fabric texture

    CN101976440A

  • Fabric defect detection method based on sparse representation coefficient optimization

    CN104778692A