The Research Of Detection And Recognition For Fabric Defects Based On Wavelet Domain Features

Fabric defect detection and recognition based on image processing approach is a key component in textile production process.Artifical offline is used in traditional detection and recognition, often affected by subjective factors, fabric defect detection and identification based on computer vision has become a research hotspot of scholars from various countries. Fabric defect is singular signal, wavelet transform provides a multiresolution framework, has good time-frequency characteristics, so it has theoretical significance and application value to discuss defect detection and recognition in wavelet domin. Adaptive denoising, segmentation, edge detection and feature extraction of fabric defect based on wavelet feature is studied in this paper,the main research work is as follows:First, Adaptive multiple threshold denoising in wavelet domain combind with maximum entropy of fabric defect segmentation method is studied according to the characteristics of noise and wavelet coefficients distribution of fabric defects, multiscale adaptive multiple threshold denoising method can effectively filter gaussian noise and smooth texture, it also retention image edge profile and other major information, and the maximum entropy segmentation method can get the most original image information, The combined method can effective segment fabric defects. Simulation contrast the largest categories segmentation method of variance , verify the validity of this method.Second, Wavelet can describe edge only in three direction,a method of fabric defect edge detection based on contourlet transform is proposed. Firstly, decompose fabric defect image on three layer by Contourlet transform, use adaptive threshold to denoise the high frequency sub-band coefficients in eight directions, then find maximum modulus coefficients in each scale and direction ,finally obtained edge of fabric defects through inverse transform ,then refinement though histogram statistics and remove isolated points. Simulation results compared with the method of edge detection based on gaussian dual wavelet, verified the fabric edge detection based on Contourlet transform can extracted a richer, better continuity edge defect results.Third, Feature curve based on wavelet coefficients is easily influenced by cycle noise, can not effectively extract features and locate defect area, a new feature extraction and recognition method of fabric defect based on wavelet domain difference coefficient is proposed. Subtract the horizontal and vertical high-frequency coefficient with smooth coefficient, then extract parameter of the feature curve effectively, then classification and recognition those characteristics with support vector machine. Experiment on normal, double end, double pick, lack weft, lack pick and hole fabric defect,simulation show the method can effectively find and segment defect area, classified recognition rate increased by 6.6 and 3.3 percent contrast with traditional image grey features and characteristics of wavelet coefficients.