Singular value decomposition for texture defect detection in visual inspection systems

In this paper we propose an algorithm for texture defects detection, which doesn't use supervised classification. The algorithm can be simply applied in an automatic visual inspection system. For localization of texture defects we calculate features of each non-overlapping region of an image via the Singular Value Decomposition (SVD) and image processing techniques. In next step the algorithm uses the fuzzy c-means clustering (FCM) to classify each region into two clusters. Finally we define a distance between centres of defective and non-defective clusters using some threshold value chosen empirically.