Fabric Defect Feature Extraction And Detection Algorithmresearch Based On Image Processing

The development of fabric defects automatic detection system for the textile industry has important practical significance.Textile industry is one of the pillar industry of foreign trade in our country, but most of our country textile product profit is very low and low value-added, is mainly due to the proportion of high quality textile products are too little, and the quality of textile products and is closely related to the fabric defects.At present most of the textile enterprises in our country belong to small and medium-sized enterprises, artificial sampling method were used to detect fabric defects, the improvement of textile product quality has been restricted.This aspect have a lot of research both at home and abroad, there are some products already in the market, but the price is too expensive, and have a low cost performance, so the popularization utilization rate is low.This paper content mainly for software testing part of the system.According to the sample of fabric defect images taken, based on the image of fabric defect feature extraction and detection algorithm is studied.First,Extracted the characteristics of the fabric image based on fractal,multifractal theory, analysis of Fourier transform and wavelet transform analysis, studied the degree of differentiates between the normal fabric samples and defect samples.And then focus on Fourier transform and wavelet transform to get a high dimensional feature of principal component analysis method was used to dimension reduction, respectively select out the characteristics of the effective combination,then up and other features in the new low dimensional feature space.And compared the result with experiment which did not reduce the dimension of characteristics.This paper introduced clustering algorithms. Due to the effect of the general FCM clustering algorithm for non spherical data set are bad, we choose fuzzy clustering algorithm based on redundancy of ideas of spectral clustering algorithm for clustering, and then distinguish detect flaw and normal samples.Through the experimental in this paper, the results show that the methods in this paper is effective.