Abstract: Focusing on identification of different types of defects on different fabrics, a defects detection method based on matrix singular value decomposition (SVD) was presented. Firstly, a region of interest (ROI) containing the defect is identified by a proposed adaptive partitioning technique. The ROI portion of fabric image is then divided into several small non-overlapping sub-images for the singular value decomposition. Since singular values are related to energy information of the image, the remaining singular values are used to restructure the sub-image by getting rid of the singular values represented the fabric texture background energy information, thus improving the energy difference between defect region and texture background. When these sub-images are used to restore the ROI area, there will be a situation that the gap is not fully connected. Binarization threshold processing is then used to eliminate the impact, thus accomplishing the fabric defect detection. Experiments have shown that the improved singular value decomposition technique presented is short time-consuming and high efficiency. Most defects can be able to identify their location and shape in the selection of the seven different fabric textures.

Key words: singular value decomposition, fabric defect detection, self-adaptive partitioning technique, ROI region