Yarn-Dyed Fabric Defect Detection based on Multi-resolution Global and Local Saliency

Abstract

In order to detect the defection of yarn-dyed fabric, a method of integrating the global and local saliency maps of multi-resolution is proposed. Multi-scale images were obtained by haar wavelet transform, and different resolutions of images were calculated global comprehensive saliency. Then GBVS algorithm was used to calculate local saliency of fabric images. The global and local saliency maps were weighted and fused to obtain comprehensive saliency images. Finally, image segmentation and morphological operations were carried out to detect the defect areas. Experimental analysed that the detection success rate of different types of texture patterns under five different kinds of defects. The experimental results showed that the detection success rate is 93.5%, so the detection rate is fast which has a certain feasibility for industrial production.