A Detection Method For Textile Image With Strip Fabric Defects

Textile flaw detection is one important field, using computer machine vision technology. The introduction of machine vision technology makes textile defects' detection operate much more efficiently, judging what done through original human hand. Now commonly used textile's defect detection methods can be concluded to two methods, namely the space domain method and the frequency domain method, both of which are macro analysis methods. Specially, the gray characteristic method, the morphological metric method, the wavelet analysis method and frequency filtering method are the main research methods.Through experimental observation, we have found that characters'values of parts with defects of the original image are different from the normal ones of the texture information. For the plain weave fabric without dyeing, we can find out some of the characteristics' attribute information through some image processing experiments. Specifically, for the defective parts, the energy characteristic's value is much bigger than other parts, the correlation characteristic's value bigger with the entropy characteristic's value smaller and the contrast characteristic's value smaller. We can use four characteristics of GLCM in textile flaw's detection by selecting the appropriate threshold. In addition, the fractal dimension characteristic's value of defective parts is much smaller, with the local entropy characteristic's value much bigger. We can also make good use of them to distinguish defective parts from normal parts, through selecting an appropriate threshold.Based on these six characteristics, we can design an algorithm to find out the fabric's strip defects. Before using these features, we use the fractional differential method to do the image preprocessing, and then we divide the original fabric image into small pieces with the same size. Then we calculate the six characteristics through the GLCM for each piece. Combining with the fractal dimension characteristic's value of each piece and the local entropy characteristic's value, we can obtain six characteristics' attribute information through a number of experimental image processing. After that, we can automatically get the appropriate thresholds. The Box-plot method used in this paper is simple and effective for strip fabric defect's detection, combining with majority voting.This paper proposes a method for textile with strip defects. In addition, the method this paper gives costs less time. Using the method, we can obtain ideal detection effect. Because threshold values are automatically extracted, the method has enough self-applicability. Using fractional differential method allows us to improve the original image, and good image quality can be achieved by reasonably decreasing the proportion of the noise information, which allows us to get better effect of detection, using some characteristics in the actual image processing.However, if the defect distribution is too complex, it is difficult for the proposed method to play a role. So, we need to improve this method to make it possible to detect much more defects with complex shape. Specifically, we can introduce more characteristics or can adopt a more excellent method for data training, to find a more appropriate threshold. For data classification, we can also use support vector machine or the method of BP neural network to train the appropriate threshold. This will be our focus of the work in the future.This article has confirmed that the presented defect detection method perform well, through a large number of experimental results and data analysis.