An adaptive fabric defect segmentation method based on a simplified PCNN

摘要

Fabric defect segmentation from an image has been proved to be a difficult task due to the influence of environment (e. g. illumination and noise, etc.) and various kinds of weave textures. Based on the property of variation with weft or warp direction in woven fabric defect, fabric images are described with the features whose variability degrees of gray contrast gradient of fabric image for segmentation relative to that of no-defective fabric image are first extracted before segmentation process. Then a simplified Pulse Coupled Neural Network (PCNN) with the parameters determined by spatial distributing information and feature data from an image is applied to adaptively segment the images. Experimental results show that the proposed method can segment common fabric defects quickly and correctly.