Real-time textural defect detection based on label run length co-occurrence matrix

A new algorithm based on label run length co-occurrence matrix is proposed for real-time textural defect detection. Standing on the real-time constraint, 256 gray-levels are reduced to several tone labels so that the computational complexity of feature extraction is greatly reduced. Several features which emphasize continuous and directional property of the defect are extracted from the label run length co-occurrence matrix and used to characterize the presence of the defect. The extracted features of testing images are subjected to a Mahalanobis distance classifier, which is built with non-defective fabric images in the learning stage. The detection results of defects in both plain and twill textured fabrics are presented to validate the proposed algorithm.