Fabric defects detection via visual attention mechanism

Because a large number of fabric defects have the characteristics of low contrast and unclear, manual detection is very tedious and inefficient. Therefore, it is necessary to detect defects automatically. An improved fabric inspection method, inspired by visual attention mechanism computation model is presented. The first step is to input fabric images with defects, and filter them by linear low-pass filtering, and extract multiscale features. The second step is to use the central difference operation to get the sub saliency map, also known as the conspicuity map, In the third step, the final saliency map is obtained by combining the conspicuity maps Finally, the most significant defect region is determined by the competition between the saliency maps.