CN114792318B

CN114792318B - Method and system for eliminating moire of textile based on image processing

Info

Publication number: CN114792318B
Application number: CN202210714388.6A
Authority: CN
Inventors: 陈小妹
Original assignee: JIANGSU YONGYIN CHEMICAL FIBER CO Ltd
Current assignee: JIANGSU YONGYIN CHEMICAL FIBER CO Ltd
Priority date: 2022-06-23
Filing date: 2022-06-23
Publication date: 2022-09-02
Anticipated expiration: 2042-06-23
Also published as: CN114792318A

Abstract

The invention relates to the technical field of textile defect detection, in particular to a method and a system for eliminating moire fringes of a textile based on image processing. The method can eliminate Moire fringes in different degrees by multiple times of focusing, and obtain multiple focusing images. And performing up-sampling on the textile image according to the focusing result to obtain an amplified image. The first normal image is obtained by a difference between the magnified image and the focused image. And eliminating the existing moire information by utilizing the pixel information of the first normal image to obtain a second normal image. And performing down-sampling on the second normal image to restore the size to obtain a third normal image, and performing matching and splicing on the third normal image to obtain a fourth normal image. According to the method, Moire fringes in different degrees are eliminated through multiple times of focusing, and the eliminated images are spliced, so that the Moire fringes in the textile image are eliminated.

Description

Method and system for eliminating moire of textile based on image processing

Technical Field

The invention relates to the technical field of textile defect detection, in particular to a method and a system for eliminating moire fringes of a textile based on image processing.

Background

And (3) detecting defects of the textile in the production process, and screening qualified products to perform the next procedure. For textile defects, the defect information on the textile is visually obvious, so that the defect detection can be realized through the information in the textile image.

In the prior art, a machine learning method is often used for building a neural network or extracting features in a textile image to detect defect information in the textile image. However, textile images contain abundant texture information, moire patterns can be formed on holes formed among the textures, the moire patterns can reduce the image quality, and characteristics in the images are affected, so that the defect detection is affected.

Disclosure of Invention

In order to solve the above technical problems, the present invention aims to provide a method and a system for eliminating moire fringes of textiles based on image processing, and adopts the following technical solutions:

the invention provides a method for eliminating moire of a textile based on image processing, which comprises the following steps:

acquiring a textile image; detecting whether a moire area exists in the textile image;

if the Moire region exists, the focal length of the camera is adjusted to be larger for multiple times according to a preset focusing step length, and multiple focusing images are obtained; carrying out up-sampling operation on the textile image according to the focusing step length to obtain a plurality of amplified images; the magnified image is equally large as the corresponding focused image; subtracting the amplified image from the focusing image and carrying out binarization to obtain a difference binary image; multiplying the difference binary image and the amplified image to obtain a first normal image;

obtaining a texture edge in the first normal image; using a connected domain formed by the texture edges as a texture hole; taking the area ratio of each texture hole to the smallest texture hole as a reference index; when the reference index is larger than a preset reference threshold value, removing the region corresponding to the smallest texture hole, reselecting the smallest texture hole to calculate the reference index until the reference index is not larger than the reference threshold value, and obtaining a second normal image;

carrying out down-sampling operation on the second normal image to obtain a third normal image which is as large as the textile image; matching and splicing a plurality of third normal images to obtain a fourth normal image; the fourth normal image is the same size as the textile image.

Further, the detecting whether the moire area exists in the textile image comprises:

and inputting the textile image into a pre-trained semantic segmentation network to obtain the moire area.

Further, the upsampling the textile according to the focusing step size comprises:

and performing the upsampling operation by using a Laplacian pyramid algorithm.

Further, the obtaining the texture edge in the first normal image comprises:

processing the first normal image according to corresponding preset clustering radii under different focal lengths by a density clustering algorithm to obtain a plurality of clustering clusters; stopping clustering when the connected domain among the clustering clusters reaches the minimum; each of the cluster clusters is the texture edge.

Further, the matching and stitching the plurality of third normal images includes:

dividing the third normal image into a plurality of sub-areas to be matched; and matching the sub-areas to be matched at different positions on different normal images, and splicing the successfully matched sub-areas to be matched to obtain the fourth normal image.

Further, the matching the sub-regions to be matched at different positions on different normal images comprises:

and matching the two sub-areas to be matched through a normalized cross-correlation algorithm.

Further, obtaining the fourth normal image further includes:

training a defect recognition network by using the fourth normal image; and inputting the fourth normal image corresponding to the collected textile image to be detected into the defect identification network, and outputting defect information.

The invention provides an image processing-based textile moir e elimination system, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and is characterized in that the processor implements any one of the steps of the image processing-based textile moir e elimination method when executing the computer program.

The invention has the following beneficial effects:

according to the embodiment of the invention, after the Moire region of the textile image is detected, the focusing images of the textile under multiple focal lengths are obtained. The depth of field of the image can be changed by adjusting the focal distance to be larger, so that the space between the image textures is enlarged, and the Moire generated by the image textures is clear. And further obtaining a first normal image through the difference binary image, and specifically analyzing pixel points in the first normal image to exclude a moire area. And matching and splicing the plurality of third normal images by a matching and splicing method to obtain a fourth normal image with good quality and without influence of moire.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart of a method for eliminating moire of a textile based on image processing according to an embodiment of the present invention.

Detailed Description

To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description will be given to a method and a system for eliminating moire fringes of textile based on image processing according to the present invention, with reference to the accompanying drawings and preferred embodiments, the detailed implementation, structure, features and effects thereof. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The following describes a specific scheme of a method and a system for eliminating moire fringes of textiles based on image processing in detail with reference to the accompanying drawings.

Referring to fig. 1, a flowchart of a method for eliminating moire on a textile based on image processing according to an embodiment of the present invention is shown, where the method includes:

step S1: acquiring a textile image; and detecting whether the textile image has a moire area.

The textile is flatly laid on a detection table, and a zoom camera is arranged above the textile for image acquisition to obtain an image of the textile. The visual angle of the camera is a forward overlooking visual angle, and the shooting position is moderate, so that multiple focusing can be realized in the focusing process subsequently, and the condition that the whole camera visual field is full of textile information after a small number of focusing times is avoided.

In the embodiment of the invention, after the textile image is obtained, in order to make the characteristic information more obvious, the textile image is grayed by adopting a weighted graying method, then the image is subjected to sliding window denoising by using a 3 x 3 Gaussian core, and noise points in the image are removed, so that the characteristic information is obvious.

And detecting whether the textile image has a moire area. Moire is a texture formed by the spatial frequency between the textures of the textile and the spatial frequency of a photosensitive element of a camera, and has a large difference with the texture characteristics of the textile, so that Moire pixel points can be identified through a semantic segmentation network, whether a Moire region exists or not is judged, and the specific training method of the semantic segmentation network comprises the following steps:

(1) and taking the textile image containing the moire pixel points as a training image. And marking the Moire pattern pixel point as 1 and other pixel points as 0 on the training image to obtain marking data. The training images and the annotation data constitute training data, eighty percent of which is used as a training set, and the remaining twenty percent is used as a test set.

(2) The semantic segmentation network adopts an encoding-decoding structure. And (4) sending the training data into a semantic segmentation encoder, extracting image features and obtaining a feature map. And the semantic segmentation decoder performs sampling transformation on the feature map and outputs a semantic segmentation image. The semantic segmentation image comprises the category of the pixel points. And judging whether a Moire region exists according to the category of the pixel points.

(3) And training the semantic segmentation network through a cross entropy loss function.

Step S2: if the Moire region exists, the focal length of the camera is adjusted to be larger for multiple times according to a preset focusing step length, and multiple focusing images are obtained; carrying out up-sampling operation on the textile image according to the focusing step length to obtain a plurality of amplified images; and subtracting the amplified image from the focusing image and carrying out binarization to obtain a difference binary image.

The focal length of the camera is increased for multiple times by presetting the focusing step length, so that the resolution of the image can be changed, the depth of field of the image is reduced, the distance between textile textures in the image is increased, and the effect of eliminating partial moire fringes is achieved.

In the embodiment of the invention, the focusing step is set to be

And a plurality of focusing images are obtained by increasing the focal length of the camera for a plurality of times, the focusing images can be represented by a sequence, and different positions on the sequence represent different focal lengths.

Because the focused image is obtained by increasing the camera focal length, the textile area in the image is larger than the size of the original textile image. And the size of the focusing image can be obtained according to the focal length after the focusing step length is adjusted, and the textile image is amplified through the up-sampling operation to obtain a plurality of amplified images. The magnified image and the focused image are equal in size and correspond one to one.

Preferably, a laplacian pyramid algorithm is used for performing upsampling operation, when the laplacian pyramid algorithm performs upsampling on a textile image, a prediction residual error is obtained in the laplacian pyramid through a difference image of adjacent layer images, and interpolation is performed on an upsampling amplification part through the prediction residual error value, so that the quality of an upsampled amplified image is ensured, and image texture feature information is kept.

And subtracting the amplified image from the corresponding focusing image, and performing binarization processing to obtain a difference binary image. In the embodiment of the present invention, the binarization processing includes: and subtracting the images to obtain a difference image, marking the point which is not 0 in the difference image as 0, and marking the pixel point which is 0 in the difference image as 1. Namely, the pixel point marked as 0 represents the moire area in the textile image, and the pixel point marked as 1 represents the normal area in the textile image. And multiplying the difference binary image and the amplified image to obtain a first normal image.

Step S3: obtaining a texture edge in a first normal image; using a connected domain formed by texture edges as a texture hole; taking the area ratio of each texture hole to the smallest texture hole as a reference index; and when the reference index is larger than a preset reference threshold value, removing the area corresponding to the minimum texture hole, reselecting the minimum texture hole to calculate the reference index until the reference index is not larger than the reference threshold value, and obtaining a second normal image.

Because the normal region in the first normal image may still have moire information that cannot be eliminated, the first normal image needs to be continuously verified, and whether a moire image exists or not is detected and removed. The verification process of the first normal image needs to be refined and detected according to image features, so that the semantic segmentation network is not applicable, and needs to be analyzed according to texture edge information in the first normal image, which specifically includes:

and processing the first normal image according to the corresponding preset clustering radii under different focal lengths by using a density clustering algorithm to obtain a plurality of clustering clusters. The cluster represents the collection of texture pixel points, a connected domain formed by texture edges formed by the texture pixel points is a texture hole, and in order to ensure that the obtained texture hole does not contain edges of adjacent textures, the clustering is stopped when the connected domain between the cluster reaches the minimum. Each cluster is a texture edge, and a connected domain formed by the texture edges is a texture hole.

In the embodiment of the invention, density clustering is performed by taking the central point of the first normal image as an initial point. The clustering radius is initially set to be 1, corresponding setting is carried out according to different focal lengths, and the clustering radius of the corresponding first normal area image is adjusted by multiplying 1.5 when the focal length is increased.

In the textile image containing the moire, the moire area fills or covers the texture holes, so that the detected texture holes are small. Therefore, the area ratio of each texture hole to the smallest texture hole is used as a reference index, when the reference index is larger than a preset reference threshold value, it is indicated that the current smallest texture hole area is covered by the Moire pixel, and the area is removed, that is, the pixel value is set to 0. And reselecting the minimum texture hole calculation reference index until the reference index is not greater than the reference threshold value, and obtaining a second normal image. It should be noted that, since the information of the region including the moire information in the first normal image is less than that of the normal region, the calculation process can be ended after a limited number of times of calculation of the reference index. The second normal image includes normal area pixel information of the textile and area information having a pixel value of 0.

Step S4: carrying out down-sampling operation on the second normal image to obtain a third normal image which is as large as the textile image; and matching and splicing the plurality of third normal images to obtain a fourth normal image.

And performing downsampling operation on a plurality of second normal images corresponding to the plurality of focusing images, recovering the image size and obtaining a third normal image. The down-sampling operation is also operated through a Laplacian pyramid algorithm, the size of the second normal image is recovered, the third normal image and the textile image are made to be as large as each other, and further, the information missing in the second normal image is processed through an interpolation algorithm similar to the up-sampling operation, so that the quality of the images is guaranteed.

The multiple third normal images correspond to focusing images with different focal lengths, so that the removal degree of moire fringes is different, all the third normal images need to be combined for matching and splicing, and a fourth normal image which is finally high in quality and is a normal area is obtained, and the method specifically comprises the following steps:

dividing the third normal image into a plurality of sub-areas to be matched; and matching the sub-areas to be matched at different positions on different third normal images, and splicing the successfully matched sub-areas to be matched to obtain a fourth normal image. In the embodiment of the present invention, the size of the sub-region to be matched is set to 3 × 3.

Preferably, the matching algorithm matches the two sub-regions to be matched by using a normalized cross-correlation algorithm. The normalized cross-correlation algorithm calculates the matching degree of the two sub-areas to be matched through a matching function, wherein the matching function comprises the following steps:

wherein,

in order to achieve a degree of matching,

for the first sub-area to be matched

The pixel value of the pixel at the location,

representing the mean of the pixels of the first sub-region to be matched,

representing a second sub-area to be matched

The pixel value of the pixel at the location,

representing the mean of the pixels of the second sub-region to be matched.

The fourth normal image is the image of the textile image without the moire pattern. The defect information can be accurately obtained by detecting the defect information of the fourth normal image, and the influence of moire on detection is avoided. Preferably, after the fourth normal image is obtained, the defect recognition network is trained by using the fourth normal image. And inputting the fourth normal image corresponding to the collected textile image to be detected into a defect identification network, and outputting defect information. The defect identification network includes:

(1) and taking a plurality of fourth normal images as training data. And marking the defective pixel point as 1 and other pixel points as 0 on the training data to obtain marking data.

(2) The defect identification network adopts a coding-decoding structure, the defect identification encoder acquires the characteristics of input data, and a defect characteristic diagram is extracted through convolution and pooling. And performing sampling transformation on the deconvolution of the defect feature map by a defect identification decoder, and outputting a defect image containing defect information.

(3) And (4) sampling a cross entropy loss function for training.

In summary, the embodiment of the invention realizes moire elimination in different degrees through multiple times of focusing, and obtains multiple focusing images. And performing up-sampling on the textile image according to the focusing result to obtain an amplified image. The first normal image is obtained by a difference between the magnified image and the focused image. And eliminating the Moire information by utilizing the pixel information of the first normal image to obtain a second normal image. And performing down-sampling on the second normal image to restore the size to obtain a third normal image, and performing matching and splicing on the third normal image to obtain a fourth normal image. According to the embodiment of the invention, Moire elimination of different degrees is realized through multiple times of focusing, and the eliminated images are spliced, so that Moire elimination in the textile image is realized.

The invention also provides a system for eliminating the moire of the textile based on the image processing, which comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein when the processor executes the computer program, any step of the method for eliminating the moire of the textile based on the image processing is realized.

It should be noted that: the sequence of the above embodiments of the present invention is only for description, and does not represent the advantages or disadvantages of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. An image processing based method for eliminating moire of textile, characterized in that the method comprises:

carrying out down-sampling operation on the second normal image to obtain a third normal image which is as large as the textile image; matching and splicing the plurality of third normal images to obtain a fourth normal image; the fourth normal image is the same size as the textile image; the upsampling operation and the downsampling operation are both operated by utilizing a Laplacian pyramid algorithm.

2. The method for eliminating moire fringes based on image processing as claimed in claim 1, wherein said detecting whether moire areas exist in said image of textile comprises:

3. The method according to claim 1, wherein the obtaining the texture edge in the first normal image comprises:

4. The method for eliminating moire fringes based on image processing as claimed in claim 1, wherein said matching and stitching a plurality of said third normal images comprises:

dividing the third normal image into a plurality of sub-areas to be matched; and matching the sub-areas to be matched at different positions on different third normal images, and splicing the successfully matched sub-areas to be matched to obtain the fourth normal image.

5. The method for eliminating moire fringes of textile based on image processing as claimed in claim 4, wherein said matching sub-regions to be matched at different positions on different said third normal image comprises:

6. The method for eliminating moire fringes based on image processing as claimed in claim 1, wherein said obtaining a fourth normal image further comprises:

7. An image processing based textile moir e elimination system comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor when executing the computer program implements the steps of the method according to any one of claims 1 to 6.