Research Of Fabric Defect Detection Based On Vision Saliency

Fabric defects affect the quality of fabric seriously.It is effective to automatic detect fabric defects which has become one of the key to quality evaluation of textile.Because of the complex of fabric texture,it takes a large challenge to defect detection based on machine vision.In recent years,there has been introduced many methods of image processing and pattern recognition based visual saliency by simulating the visual attention mechanisms of human,and has achieved good results.In this thesis,we conduct intense research on fabric defect detection based on visual saliency.For the complex texture characters of fabrics,in view of its directivity and randomness,we have proposed several effective fabric defect detection algorithm by simulating the hierarchical processing mechanism of human visual perception pathway.By combining background estimation and wavelet transform or stationary wavelet transform,we proposed a novel fabric defects detection algorithm based on stationary wavelet transform and background estimation.First,for given fabric images,extract feature map by wavelet transform or stationary wavelet transform.Second,segment feature map to obtain multiple background images and then obtain multiple sub-salience maps by calculating Euclidian distance between the original image and every background image,and get saliency map which contains candidate defects regions by fusing multiple sub-salience maps.Finally,acquire defects saliency map by the Gaussian distribution model based on global estimation,and get the final detecting results by threshold segmentation algorithm.Experimental results show that the algorithm can accurately locate the defect regions,and stationary wavelet transform has a higher accuracy than state-of-the-art methods.We propose a fabric defects detection method based on mutual information measure and context analysis by using the basic property of the mutual information.First,the fabric image is divided into image patches with the same size.Then,calculate information entropy between each patch and K-image patches in the surroundings.Second,calculate the similarity based on contextual analysis to generate visual saliency map.Finally,locate defects by threshold segmentation algorithm.Experimental results show that this algorithm has a good performance and it is better than other state-of-the-art fabric defects detection algorithms existing currently.