Fabric defect detection using adaptive wavelet

This paper studies the adaptive wavelet design for fabric defect detection. In order to achieve translation invariance and more flexible design, the wavelet design focuses on the nonsubsampled wavelet transform. We design the wavelet filters under the constraints that the analysis filters are power-complementary and the wavelet has only one vanishing moment, which corresponds to a multiscale edge detector. Based on lattice structure factorization, the design of the power-complementary filters turns out to be unconstrained optimization of lattice coefficients. Adaptive wavelets are designed for five kinds of fabric defects in the experiments. Comparing the proposed method with adaptive wavelet design for defect detection based on orthogonal wavelet transform energy between the defect area and the background, and achieve a robust and accurate detection of fabric defects.