Automatic Detection of Texture Defects Using Texture-Periodicity and Gabor Wavelets

In this paper, we propose a machine vision algorithm for automatically detecting defects in textures belonging to 16 out of 17 wallpaper groups using texture-periodicity and a family of Gabor wavelets. Input defective images are subjected to Gabor wavelet transformation in multi-scales and multi-orientations and a resultant image is obtained in L2 norm. The resultant image is split into several periodic blocks and energy of each block is used as a feature space to automatically identify defective and defect-free blocks using Ward's hierarchical clustering. Experiments on defective fabric images of three major wallpaper groups, namely, pmm, p2 and p4m, show that the proposed method is robust in finding fabric defects without human intervention and can be used for automatic defect detection in fabric industries.