Real Time Textile Defect Detection Using GLCM in DCT-Based Compressed Images

This paper proposes a framework for textile defect detection on the basis of energy and contrast features of the graylevel co-occurrence matrix (GLCM). We calculate these features in DCT-based compressed image (DCTb-I), which was produced by using only a small number of DCT coefficients. The proposed detection approach complies with the Moving Picture Experts Group (MPEG) and Motion Joint Photographic Experts Group (MJPEG) video compression standards, which improves the defect detection efficiency, in terms of space and computational cost by replacing the frame grabbers with MPEG encoders and by only considering I-frames in the MPEG or specific frames in the MJPEG sequence. We determine (1) the detection accuracy of using DCTb-Is with various compression rates, and (2) the lowest DCTb-I rates that allow defect detection. The DCTb-Is are tiled into blocks from which the energy and contrast features are calculated. The highest energy and lowest contrast blocks are detected and located in the textile as defects. The proposed approach is suitable for real-time detection because, unlike common defect-detection approaches, prior feature extraction from reference images is not required. Simulation results suggest that holes, stains and drop stitches can be detected and located from DCTb-Is if the compression rate is greater than 25%.