Warp-knitted Fabric Defect Segmentation Based on the Shearlet Transform

Abstract

The Shearlet transform has been a burgeoning method applied in the area of image processing recently which, differing from the Wavelet transform, has excellent properties in processing singularities for multidimensional signals. Not only is it similar to the performance of the Curvelet transform, it also overcomes the disadvantage of the Curvelet transform with respect to discretization. In this paper, the Shearlet transform with segmented threshold de-nosing is proposed to segment a warp-knitted fabric defect. Firstly a warp-knitted fabric image of size 512*512 is filtered by the Laplacian Pyramid transform and decomposed into low frequency and high frequency coefficients. Secondly the high frequency coefficients are operated with a pseudo-polar grid and then convoluted by the window function. Thirdly the shearlet coefficients will be obtained through redefining the Cartesian coordinates from the pseudo-polar grid coordinates and de-noised by the segmented threshold method. Then the coefficients which have high energy are selected for reconstruction in an inverse way using the previous steps. Finally the iterative threshold method and object operation based on morphology are applied to segment out the defect profile. The experiment’s result states that the Shearlet transform shows excellent performance in segmenting a common warp-knitted fabric defect, indicating that the segment results can be applied for further defect automatic recognition.