Warp-Knitted Fabric Defect Segmentation Based on Non-Subsampled Wavelet-Based Contourlet Transform

Abstract

In this paper, a non-subsampled wavelet-based contourlet transform (NWCT) is applied in warp-knitted fabric defect segmentation. Compared with the traditional contourlet transform, wavelet transform takes the place of Laplacian pyramid in NWCT and the directional filter bank is non-subsampled. The wavelet transform with improved wavelet threshold is put to use, and the original fabric image can be decomposed into low-frequency approximate coefficient A and high-frequency detail coefficients V, H, and D. The high-frequency detail coefficients are processed by the non-subsampled directional filter bank to get directional sub-band coefficients. Afterward, the effective sub-band coefficients based on regional energy are chosen to reconstruct V, H, and D. And the reconstructed fabric image will be achieved by inverse non-subsampled wavelet-based contourlet transform. The adaptive threshold method and morphological processing are used to obtain the legible defect profile. The experiment demonstrates that NWCT can achieve the positive segmentation regarding the common defects, such as broken warp, width barrier, and oil, and has excellent performance on these directional defects and regional defects. It is acknowledged that NWCT will provide a new way to detect warp-knitted fabric defects automatically.