Research Of Defect Segmentation Method Based On Fabric Surface Texture

Fabric defect segmentation is the key of the automatic detection system of fabric defect, which directly affects the defect classification and recognition results. Most of existing defect segmentation algorithms is only applicable to the defect image which high contrast and the background texture not saliency. For low contrast, background texture saliency defect image is difficult to adapt. Aiming at this problem, research of defect segmentation method based on fabric surface texture. So, the research subject has important practical significance and application value.After reading lots of literature and studying the fabric defect segmentation. We research the defect segmentation method for fabric image which has low contrast and background texture saliency. In this paper, the main work and research results are as follows:(l)Study on the different extraction of normal fabric texture and defect texture. Analysis of fabric visual saliency features. Present a method to describe the local texture pattern of coarseness, direction and contrast extraction. Experiments show that, the extracted texture coarseness feature map, the texture directional features map and the contrast feature map to highlight the defect region.(2)In view of the existing local texture coarseness algorithm selected the neighborhood size, using the linear quantitative accuracy is not high. The improved local texture coarseness algorithm is proposed, which is efficient proved by experimental results.(3)According to the extraction of the fabric texture is saliency. A method is presented for the texture saliency feature extraction algorithm. Firstly, calculate each pixel corresponds to the best window size by used improved local texture coarseness algorithm. Secondly, coarseness, direction and contrast are extracted from the best windows of fabric image. Finally, they are normalized and a texture salient feature map is generated by using weighted fusion. The proposed algorithm has good robustness in the experiments of TILDA database and collected images by CCD.(4) For traditional pulse coupled neural networks model is too complex and numerous parameters. The artificial selection of network parameters is needed in image segmentation. An improved PCNN defect segmentation algorithm is proposed and according to the information of the image itself adaptive setting the parameters of the improved PCNN model. We apply the approach to texture salient feature map, the experimental results show that it is efficient.