Research on fabric defect detection via different filters combination in NSCT

A new kind of NonSubsampled Contourlet Transform (NSCT) based on 9/7wavelet filter banks with rational coefficients is given in this paper. And four kinds of NSCT methods, one of which is this new NSCT, are applied to a new fabric defect detection algorithm respectively. In the algorithm, the fabric defect image is firstly decomposed into different frequency subbands by NSCT. Secondly, both the low frequency subband and the optimal high fequency subbands selected from all high frequency subbands by using a cost function are thresholded. Finally, these thresholded subbands are fused and then binarized in order to separate the defect from the image texture background. The contrast experiment results of four kinds of NSCT methods show that this new NSCT has good performance both in defect detection effect and in algorithm´s execution time.