Warp-knitted fabric defect segmentation based on non-subsampled Contourlet transform

Abstract

Differing from the traditional Contourlet transform, the non-subsampled Contourlet transform (NSCT) is proposed to apply in warp-knitted fabric defect segmentation. First, the Laplacian pyramid is used to achieve the decomposition of original fabric image. Second, the high frequency directional sub-band coefficients are extracted by means of the non-subsampled directional filter bank. Then, choose the best high frequency sub-band coefficient at every level based on regional energy maxima and reconstruct the image. Finally, the legible defect profile is obtained by adaptive threshold method and morphological processing. The experimental results including the common defects, such as broken warp, width barrier and oil, show that the NSCT could attain the correct segmentation on directional defect and regional defect. This method fits the directional changes of warp-knitted fabric defect. It provides a new way to detect warp-knitted fabric defects automatically.