中文摘要：

为了检测混凝土表面裂缝及其宽度，对含有裂缝的数字图像进行k-均值聚类，提取出所有疑似裂缝的像素点，进行了二值化处理。根据像素点的位置关系，提取连通分量，将连通分量作为聚类对象，构造连通分量间的距离函数。利用谱聚类算法将连通分量聚类，根据裂缝特征，去掉伪裂缝部分，得到完整的裂缝对象，并通过局部旋转算法对裂缝的宽度进行了2次数值计算。分析结果表明：与Canny、Sobel算子比较，多级聚类算法在裂缝提取时能去掉较多的噪声，抗噪能力强；通过局部旋转算法计算裂缝宽度时，计算值与实际值的平均相对误差分别为3．86％、2．40％，算法精度高，适用于各种类型裂缝宽度计算。

英文摘要：

In order to detect the crack and its width of concrete surface, the k-means clustering was applied for crack digital image, and binary image was got based on taking out entire suspected crack pixels from clustering results. The connected components of binary image were extracted according to the ubiety of pixels, the distance function of connected components was constructed considering connected components as clustering objects. Connected components could be clustered by spectral clustering algorithm, pseudo cracks were removed on the basis of crack features, and whole crack image was obtained. Numerical calculations of crack width were carried out twice by local rotation algorithm. Research result shows that multi-level clustering algorithm can get rid of more noises during extracting crack, and has stronger anti-noise ability compared with Canny operator and Sobel operator. When the crack width is calculated by local rotation algorithm, the average relative errors of calculated value and actual value are 3.86% and 2.40% respectively, so the algorithm has high accuracy and can be used for width calculations of all kinds of cracks. 3 tabs, 9 figs, 14 refs.