Texture Defect Detection with Non-Supervised Clustering

In this paper a new algorithm for texture defect detection, which can be used in automatic visual inspection system, is presented. For the purpose of detect and localize texture defects it divides up texture image into non-overlapping areas. Then it applies principle component analysis (PCA) to calculate feature describing each area. Finally it uses fuzzy c-means clustering (FCM) to classify each area as defective or non-defective. Presented algorithm was used for the defect analysis in sample defective and non-defective natural textures. Experimental results proved that proposed texture defects detection method is effective for real texture surface.