Crack Detection Using Fast Spectral Clustering Considering Graph Connectivity

Cracks on pavement roads cause various traffic problems. Therefore we should repair them properly. Nowadays a variety of crack detection methods in computer vision have been proposed. Spectral clustering is one of them and an effective method, but suffers from processing time due to the large amount of calculation. Among them, calculating of Laplacian-matrix and eigenvalues especially affect processing time. Therefore we propose two methods to improve the efficiency of algorithm. One applies sparse process considering graph connectivity for Laplacian-matrix, and the other considers amount of pixel of crack of pavement road images.