Surface Crack Detection Algorithm Based on Double Threshold and Tensor Voting

The double threshold method based on the combination of bilateral filter and local grayscale difference is proposed to extract crack segments, and tensor voting algorithm is adopted to solve the problem of crack extraction caused by low contrast between cracks and background, as well as unevenness of gray values within the crack region. The double threshold method is introduced to obtain crack segments, and then based on proximity and continuity of crack fragments, the significant map and complete center line arc obtained with tensor voting. Accurate crack extraction is realized by connecting crack fragment and removing discrete points with center line. Experimental results show that, compared with the method based on the beginning and end of crack fragments to connect, the proposed algorithm can increase F-measure about 27% to process the surface image of ceramic elements with cracks.