CN103456021A - Piece goods blemish detecting method based on morphological analysis - Google Patents

Abstract

The invention discloses a cloth defect detection method based on morphological analysis, which uses a line array camera to collect images of cloth in motion, and obtains the distance between adjacent tissue points of the cloth through the grayscale image of the cloth as a reference for the size of structural elements value; followed by a morphological closing operation to fill the gaps of conventional weaving points, the number of weaving points at the defect is relatively small and irregular, and is highlighted in brightness; finally, the image is binarized, and the threshold is determined by gray The γ training selection of the degree power transformation is controlled, the binary defect map is counted and marked, and the defect information is output to complete the defect detection process of the cloth. The structure of the method of the invention is more accurate and simple, and the calculation speed is improved; the interference of uneven background brightness can be weakened; and the automatic detection is facilitated.

Description

A cloth defect detection method based on morphological analysis

technical field

The invention relates to a fabric detection method, in particular to a cloth defect detection method, in particular to a cloth defect detection method based on morphological analysis.

Background technique

In the textile production process, quality control and testing are very important, and cloth defect detection is a particularly critical part of it. At present, the detection method of most textile enterprises is still manual detection. This method has a low detection speed and is easily affected by the subjective factors of the inspectors. False detections and missed detections often occur, and domestic labor costs are getting higher and higher. Therefore, manual detection is increasingly unable to meet the requirements of modern manufacturing enterprises. Replacing manual detection with advanced automatic detection technology can improve detection efficiency, reduce labor force, reduce labor intensity and further improve the quality of cloth.

At present, there are cloth inspection systems on the market, mainly including the Cyclops cloth inspection system of the Belgian company Barco, the I-TEX2000 cloth inspection system of the Israeli Evs company, and the Fabriscan cloth quality automatic inspection system of the Swiss Uster company. Most of the above-mentioned equipments are very expensive, and most domestic textile enterprises cannot afford them.

In the prior art, cloth defect detection methods are roughly divided into three categories: statistical methods, model-based methods, and spectral analysis-based methods. The more classic methods are: gray co-occurrence moment method based on statistics, Fourier transform method based on spectrum analysis, wavelet transform method, Gabor filter method and Markov random field method based on model. Among them, the commonly used detection algorithm is based on the method of spectral analysis or combined with other methods. These methods extract relevant feature information through the conversion between the space domain and the frequency domain, but the method involving the frequency domain has a large amount of calculation and the conversion between domains. It is difficult to guarantee the accuracy of some information.

With the development of line-scan cameras, high-speed acquisition of high-resolution images becomes possible, coupled with the advantage of the intuitiveness of the spatial domain, which provides a basis for the development of some new defect detection methods based on the spatial domain.

Contents of the invention

The purpose of the present invention is to provide a method for detecting cloth defects based on morphological analysis, so as to reduce the cost of automatic detection of cloth defects and replace the manual detection methods commonly used in China.

In order to achieve the above-mentioned purpose of the invention, the technical solution adopted in the present invention is: a method for detecting cloth defects based on morphological analysis, comprising the following steps:

(1), adopt line array camera to carry out image acquisition to the cloth in motion, obtain the cloth grayscale image that bit depth is 8;

(2), carry out basic luminance stretching and binarization processing to the gray image of the cloth, count the distribution number of weave points, and calculate the distance r between adjacent weave points of the cloth, as the reference value of the size of the structural element;

(3) Structural elements are selected according to the type of defect. For a defect that is approximately circular, the structural element is a disc structure with a radius of r; for linear defects, the structural element is a structure with a side length of r square structure;

(4) Contrast stretching is performed on the original gray image of the cloth through top and bottom hat transformation, while weakening the influence of ambient light on the image;

(5) The contrast-stretched gray-scale image is processed by means of expansion and erosion in morphology, first expanded and then corroded, to fill the dark gaps between conventional tissue points, and form a brightness contrast with the blemish;

(6), carry out the negative film operation to the image obtained in (5) with the gray scale power transformation, make the blemish part become the main body of the bright color, train and adjust the gray scale power transformation in the same time value, the gray value is stretched and compressed until the gray value of the intact part and the defective part are distributed on both sides of 127.5 under the same ambient brightness;

(7), the image obtained in (6) is subjected to binarization processing with a threshold value of 127.5, and an expansion operation with a structural element size of r is carried out to fill in the blemishes;

(8), in the image after binarization processing, for the points with a value of 1, use the eight-connected method to mark the connection, and the number of connected regions is the number of defects; count the number of pixels contained in each connected region The sum is the area of the defect, and the centroid of the connected region is calculated as the center of the defect, thereby realizing the detection of cloth defects.

In the above-mentioned technical scheme, firstly, the basic binarization operation is performed on the input image, and the distance between adjacent weaving points of the cloth is analyzed to obtain the size of the structural elements for the subsequent morphological processing; Image contrast, so that it is not limited to a small gray scale range; then the morphological closing operation is performed to fill the gaps of conventional weaving points, and the number of weaving points at the defect is relatively small and irregular, which is highlighted in brightness; finally, the The image is binarized, and the determination of the threshold is controlled by the training and adjustment of the γ value in the gray power transformation. The binary defect map is counted and marked, and the defect information is output to complete the defect detection process of the cloth.

In the above technical solution, in step (2), the number of tissue points is directly counted geometrically from the perspective of the space domain, so as to calculate the distance between adjacent tissue points as a reference value for the size of the morphological structural elements.

值代替二值化的阈值。在进行负片操作的时候进行灰度调整，灰度调整的方式为幂次变换，

、

分别为变换前后的灰度图像，则

，

、γ为正常数，

取为1，γ值由对无瑕疵图像的训练调整得到。 In step (6), training

value instead of the threshold for binarization. The grayscale adjustment is performed during the negative film operation, and the grayscale adjustment method is power transformation.

,

are the grayscale images before and after transformation respectively, then

,

, γ is a normal constant,

Taken as 1, the γ value is adjusted by training on unblemished images.

Due to the use of the above-mentioned technical solutions, the present invention has the following advantages compared with the prior art:

The present invention solves the complex practical problem of flaw detection by analyzing the image structure and adopting the method of mathematical morphology from the perspective of space domain. The difficulty lies in the determination of the size of the morphological structural elements. The selection of structural elements is the key to ensure the accuracy of morphological operations. The present invention uses a line array high-definition camera to collect images. The structure of the cloth weaving points is clear, and the structure can be calculated using geometric knowledge. The reference value of the value of the element does not need to be converted to the frequency domain to calculate the relevant scale parameters, and the structure of the method is more accurate and simple, and the calculation speed is improved; in addition, the present invention does not use general brightness transformation when performing contrast stretching. Instead, the top-hat and bottom-hat transformation methods are used, because the interference of background brightness unevenness can be weakened during the top-hat transformation; moreover, the present invention converts the setting of the threshold value during the binarization process into grayscale power transformation. The reason for the selection of the γ value is that the threshold value of different images is different because the change of ambient brightness affects the gray value of the image. If this variable is controlled from the perspective of gray level, the processing result will be more accurate.

Description of drawings

Fig. 1 is a flow chart of a method for detecting cloth defects based on morphological analysis in an embodiment of the present invention;

Fig. 2-Fig. 6 are the common sample images of cloth defects collected in different ambient brightness in the embodiment;

Figure 7-Figure 12 is the image of the defect detection process in Figure 4 as an example;

Figures 13-16 are the output images of the detection results of common cloth defect images.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing and embodiment:

Embodiment 1: Refer to FIG. 1 , which is a flow chart of research on a cloth defect detection method based on morphological analysis. Firstly, the structural elements of the collected images are calculated, the contrast is enhanced by morphological methods, the closing operation is used to fill the gaps of tissue points, the compressive negative film operation is performed, binarization is performed, and relevant information is output by statistically marking defects.

Figure 2 to Figure 6 are typical defect images collected by the present invention, and the detection algorithm is described in detail below taking Figure 4 as an example:

1. Calculate the size of the structural element

First, calculate the distance between adjacent tissue points of the collected cloth image. When the cloth image is large, sample segmentation and collection can be performed on the cloth image. For example, the intercepted length of a=500 and width of b=200 are roughly distributed in the original image. The 4 sample images in , as shown in Figure 7.

，结果作为结构元素大小。 The sample image is stretched linearly, so that the gray value of the image is covered from 0 to 255, and the threshold value of 127.5 is used for binarization, and the tissue points in the image appear as white bright spots (value 1). Use the eight-connected method to mark the connected components of the points with a mark value of 1, and the statistics show that the average number of distribution points of each sample tissue of 500×200 pixels is n=1376, then the formula for calculating the distance between adjacent tissue points of the cloth is as follows:

, the result as the structuring element size.

Structural elements are selected according to the types of common defects, which are usually divided into two types in terms of structure: one is defects such as holes and stains with high circularity, as shown in Figures 2, 3, and 5. The structural elements adopt a disc structure. The radius is taken as r; the other category is defects such as missing warp and weft, which are narrower in width and similar to linear shapes, as shown in Figures 4 and 11. At this time, the structural element adopts a square structure, and the side length of the square is taken as r.

2. Enhance the contrast of the original defect map

The contrast of the collected original image is stretched through the top-hat transformation, and at the same time, the influence of ambient light on the uneven brightness of the image is weakened. The gray value of the image captured by the camera is relatively concentrated, and the difference between the pixel value of the defect and the weaving point is not large enough, as shown in Figure 4, the gray value ranges from 47 to 186, that is, the contrast is not obvious, so the contrast should be enhanced. The commonly used method is linear brightness transformation, which stretches 47~186 to 0~255 according to a linear ratio, but observing the image shows that its brightness is not uniform, and linear brightness transformation cannot reduce the ambient brightness interference.

The present invention selects the morphological contrast stretching transformation based on top-hat and bottom-hat transformation. The top-hat transformation is the result of subtracting the opening operation from the original image. For an image with an uneven background, when the structural element takes an appropriate value, the opening operation obtains the background, and the result of the top-hat transformation is an image that excludes the influence of the background. Similarly, the bottom hat transformation is the result of subtracting the closing operation from the original image.

为灰度图像，

为结构元素。 in,

is a grayscale image,

as a structural element.

The above two transformations can be used together to enhance contrast, which is called the morphological contrast transformation.

Figure 8 shows the result of this contrast stretching for warp-missing defect images.

3. Organization point gap filling

The contrast-stretched gray-scale image is processed by dilation and corrosion, first dilation and then corrosion, filling the dark gaps between conventional tissue points, and forming a contrast in brightness with the blemishes.

，结构元素为

，则腐蚀、膨胀分别为 For grayscale images it is

, the structural elements are

, then the corrosion and expansion are respectively

和

分别是

和

的定义域，平移参数及

在的定义域内，而

和

在

的定义域内。 in,

and

respectively

and

The domain of definition, the translation parameter and

exist within the domain of definition, and

and

exist

within the domain of definition.

Corroding first and then dilating Fig. 8 can stretch upward the values with very small gray values, that is, the troughs, and make the troughs gentle. Overall, except for the blemishes, the overall brightness of other parts is brighter, and the blemishes are highlighted, as shown in Figure 9.

4. Negative conversion and binarization

Negative operation is performed on the image obtained by filling the gap with grayscale power transformation, so that the blemishes become the main body of bright color. At the same time, the γ value of the power transformation is trained and adjusted, and the grayscale value is properly stretched and compressed, so that the image under the same ambient brightness The gray values of the intact part and the defective part are distributed on both sides of 127.5.

、γ为正常数，

由训练得到,

取为1。γ<1时，映射偏重更高数值（明亮）输出；γ>1时，映射偏重更低数值（灰暗）输出。 The power transformation is a nonlinear brightness transformation, ,

, γ is a normal constant,

obtained by training,

Take it as 1. When γ<1, the mapping favors higher-value (bright) outputs; when γ>1, the map favors lower-value (dark) outputs.

Because the image shown in Figure 9 obtained after the closed operation is not only the dark color of the defect, but some pixels have little difference from the defect. In this environment, the dark color should be compressed to a narrow range, so γ<1 is selected. The specific value of γ is determined by training on flawless images. The critical value of this missing image is 0.92 after the training of the perfect image in the environment. γ should be as close to the critical value as possible. If it is too small, it will cause too much brightness compression, resulting in missed detection. Take 0.90 , the result is shown in Figure 10. As shown in Figure 2 and 3, under the ambient brightness γ<1, the dark color range needs to be compressed at this time, and the γ value is smaller than that in Figure 4; as shown in Figure 5 and 6, under the ambient brightness γ>1, the bright color range needs to be compressed at this time . Then perform default binarization, and perform appropriate expansion to fill in the blemishes and dents, and remove edge burrs, as shown in Figure 11.

5. Statistical defect information and output

After binarization, the background pixel value of the image is 0 (black), and the pixel value of the defect is 1 (white). For the points marked as 1, the eight-connected method is used to mark the connection, and the number of connected regions is the number of defects. ; The sum of the pixels contained in each connected region is counted as the area of the defect, and the centroid of the connected region is calculated as the center position of the defect. Finally, output the defect image and the above defect parameters, as shown in Figure 12.

Figures 7 to 12 intuitively show the detection process of warp-missing defects, and the detection results of other common defects are shown in Figures 13 to 16.

Claims (3)

1. A cloth flaw detection method based on morphological analysis is characterized by comprising the following steps:

(1) acquiring images of the moving cloth by using a linear array camera to obtain a cloth gray image with the bit depth of 8;

(2) carrying out basic brightness stretching and binarization processing on the gray level image of the cloth, counting the distribution number of tissue points, and calculating the distance r between adjacent tissue points of the cloth to be used as a reference value of the size of a structural element;

(3) selecting structural elements according to the defect types, wherein for the defects which are approximately round, the structural elements are of a disc structure with the radius of r, and for the linear defects, the structural elements are of a square structure with the side length of r;

(4) carrying out contrast stretching on the original gray level image of the cloth through top and bottom cap conversion, and weakening the influence of ambient light on the image;

(5) processing the gray image after contrast stretching by adopting a morphological expansion and corrosion mode, wherein expansion is performed first and then corrosion is performed to fill dark gaps among conventional tissue points and form contrast on brightness with flaws;

(6) performing negative film operation on the image obtained in the step (5) by using gray power transformation to enable the flaw part to become a bright color main body, training and adjusting a gamma value in the gray power transformation, and stretching and compressing the gray value until the gray values of the intact part and the flaw part under the same environment brightness are distributed on two sides of 127.5;

(7) carrying out binarization processing on the image obtained in the step (6) by taking 127.5 as a threshold value, carrying out expansion operation with the size of a structural element r, and carrying out recess filling on the flaw;

(8) in the image after the binarization processing, marking and communicating points with the value of 1 in an eight-communication mode, wherein the number of the communicated areas is the number of the flaws; and counting the sum of pixels contained in each connected region as the area of the defect, and calculating the center of mass of the connected region as the central position of the defect, thereby realizing the detection of the cloth defect.

2. The method of claim 1, wherein the method comprises: in the step (2), the number of the tissue points is directly counted in a geometric mode from the angle of the spatial domain, so that the distance between the adjacent tissue points is calculated and used as a reference value of the size of the morphological structure element.

3. The method of claim 1, wherein the method comprises: in the step (6), training a gamma value to replace a binary threshold value; performing gray scale adjustment during negative film operationThe whole way is a power transformation,

、

gray scale images before and after transformation respectively

，

Gamma is a normal number, gamma is a positive number,

taken as 1, the gamma value is obtained from the training adjustment of the inpainted image.

Images