The Detection Of Fabric Defects Based On Contourlet Transform

Fabric defect detection is the key link of textile quality control and management. Atpresent, the fabric surface defect detection is also largely dependent on human eyes. Thismethod can not meet the needs of modern production as the low efficiency, high falsedetection rate, strong subjectivity and other defects. Therefore, using machine vision toautomatical fabric defect detection is a hot research topic in the field of textile engineering.If we want to make use of machine vision to fabric defect detection, the primary task isto develop an effective online detection procedures. The detection procedures are generallydivided into defect image segmentation and classification. However, the fabric images will besubject to noise pollution during captured and transported. Therefore, the fabric image shouldbe preprocessed before segmentation to eliminate noise in the image. In order to eliminatenoise impact of defects feature on fabric defect images, highlight the fabric defects features,Donoho multi-scale stratification threshold which is based on Contourlet high frequencydirectional sub-band energy is innovatively proposed on the basis of previous studies forfabric defect image denoise. This method has better denoising effect than wavelet denoisingand contourlet Donoho multi-scale threshold denoising.Fabric defect segmentation which can make fabric defect visually represented is animportant part of the expression of fabric defects. On the basis of Contourlet transform,high-frequency sub-band coefficients are screened by regional energy maxima andrecombined new coefficients. Then a new image is reconstructed by the new high frequencysub-band coefficients and low-frequency coefficient. Finally, thresholding methods andmorphological opening operation are used to achieve an effective fabric defect imagesegmentation. In Contourlet decomposition section, the choice of Laplacian pyramid (LP)decomposition filter, directional filter banks (DFB) and decomposition level are described indetail to determine the optimal decomposition scheme. Under optimal decomposition scheme,the method of fabric defect segmentation based on Contourlet transform proposed in the paperis better to capture fabric defects contour information and retain its features.In order to achieve an effective fabric defect classification and get a more intuitiveunderstanding of fabric defect, feature extraction and classification of BP neural network arecombined in the paper. In this section, texture features of fabric defect images and fusedimages are extracted by GLCM. Meanwhile, calculating structure features of segmentedimage. Then screening the features uses principal component analysis. Finally, the datascreened by principal component analysis are input into a three layers BP neural network totrain it and detect fabric defect. Five common classes of fabric defects in weaving and normalfabric image are produced as experimental subjects in the paper. The results show that therecognition accuracy is up to98.3%.In order to be more intuitive to express this algorithm, Matlab GUI toolbox is used torepresent it with a single user interface.