Pavement distress detection has significant importance for maintaining and managing roads, and a variety of crack detection methods based on pavement images have been proposed. However, their accuracy is vulnerable to noise on images. In this paper, we introduce the continuity of cracks as one of the features of the cracks into a method using spectral clustering to avoid misdetection caused by noise. The continuity is introduced into affinity matrix. The result of experiment shows the accuracy gets higher by our approach.