The pavement image has three features: the background has similar statistical properties; cracks have less brightness than the background; cracks are connected to linear structure with much bigger area than impurities in the same image. Using these features, we present an algorithm comprising of image segment, breaking region connecting, and acreage comparison to extract pavement crack. Firstly, image segmentation using k-means clustering is implemented, and then the region with the minimum mean gray value is set as the initial extracted crack. The eight-connectivity cracks with small area are excluded to suppress interference and improve detection accuracy rate. Secondly, the minimum distance of every two non-connected crack regions is computed. When the distance is smaller than a given threshold, connect the two regions in the shortest possible path. Finally, morphology method is adopted to extract skeleton. Experimental results testify the effectiveness of the proposed algorithm in extracting pavement image cracks.