针对裂纹与背景之间的低对比度及裂纹区域内灰度值不均匀所导致的裂纹提取困难问题,提出一种基于双边滤波和局部灰度差相结合的双重阈值裂纹片段提取法,并结合张量投票算法进行裂纹检测。该算法采用双重阈值法获取裂纹片段,并根据裂纹片段的接近度和连续性特征,通过张量投票算法得到裂纹的显著性图谱以及完整的裂纹曲线,利用裂纹曲线对裂纹片段进行连接并去除离散点,完成准确裂纹提取。实验结果表明,相比于根据裂纹片段首尾位置进行连接的方法,该算法处理类陶瓷元件表面裂纹图像时F-measure提高了约27%。

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 are 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.