Fabric Defect Image Segmentation Based On The Building Of Visual Saliency Map

In the textile enterprise,fabric defect detection is completed by inspectors on the inspecting machine. However,this kind of detection method based on artificial vision usually cannot satisfy the needs of online detection. So, developing automatic detection equipment based on machine vision is particularly important. While,the key to developing this kind of automatic detection equipment is the researching of image processing algorithm. However, the gray- level of defect is usually close to the normal texture in fabric image, traditional method of image processing is hardly to realize the segmentation of textile image completely.In this paper, first, through analysis of the mechanism of human visual significant mechanism, on the basis of reference Itti model and Stentiford model, the Itti visual saliency is optimized and improved. Then the Wavelet transform and Fourier transform is used for building the intensity saliency map of fabric image, and a kind of method for measuring and improving the global intensity saliency of fabric image is presented, also presenting a method of measurement for local intensity saliency based on residual of the Fourier energy spectrum. Next, through analyzing the characteristics of fabric image itself, the Gabor filter group is designed and optimized for the extracting of direction characteristics, this greatly improve the efficiency of direction feature extraction and the building of the direction saliency map. Finally through segmenting the saliency map with the threshold and linear fusion, the information of defect is formed.The experimental results show that the method of this paper which through building the saliency map of fabric image improve the saliency of fabric defect, which can improve the segmentation accuracy and integrity compared with the traditional algorithm. The research results of this article will provide theoretical support for online fabric defect detection.