Abstract: Edge detection of fabric defect based on wavelet transform can get edge information only in the direction of horizontal, vertical and diagonal, likely to cause loss of directional edge. An edge detection method based on Contourlet transform is proposed. The high frequency sub-band coefficients are denoised on each direction of Contourlet transform using adaptive threshold. Maximum modulus coefficients are found in each scale and direction. Edge of defects is obtained through inverse transform of maximum modulus coefficients and low frequency coefficients. Refinement is given though histogram statistics and isolated points are removed. Simulation results show the method obtains a more detailed edge defect information. Particularly the detect results more upon to the original blemish are got.

Key words: fabric defect, multi-directional, edge detection, adaptive threshold

摘要： 基于小波的多尺度织物疵点边缘检测算法只能在水平、垂直和对角方向获取边缘信息，容易造成方向性边缘的丢失。提出一种基于Contourlet变换的多方向织物疵点边缘提取算法，对织物疵点图像Contourlet分解后多个方向上的高频子带系数进行自适应阈值去噪，求取各方向子带系数的模极大值，将高频模极大值系数与低频系数进行反变换，通过直方图统计及去除孤立点的细化方法得到织物疵点边缘。仿真结果表明，该方法得到的疵点边缘信息更加丰富，尤其对棉结等区域类疵点能得到更加逼近疵点真实边缘的检测结果。

关键词: 织物疵点, 多方向, 边缘检测, 自适应阈值