Abstract: As it is influenced by inspection environment and the characteristics of defect, the traditional detection methods are difficult to meet the requirements of defect dynamic detection. A new method based on visual saliency mechanism is presented in this paper. First of all, characteristic pattern is obtained through extracting the features of acquired image. Secondly, the characteristic sub-images are formed by wavelet multilevel decomposition of the characteristic pattern. On this basis, the center-surround difference operation is used to construct the characteristic difference sub-image. Then, through fusing these characteristic difference sub-images, the salient image is formed. Finally, extracting the region of interest use the threshold method, and then segmenting the defect targets use the region growing method. Experimental results show that this method can detect plain weave fabric defect information, and has strong anti-jamming ability.

Key words: visual saliency mechanism, feature extraction, saliency map, fabric defect detedtion