Pavement Crack Detection Using Progressive Curvilinear Structure Anisotropy Filtering and Adaptive Graph-Cuts

Abstract

Delineation of pavement cracks is essential for the damage assessment and maintenance of pavements. Existing methods are not sufficiently robust to interferences including varied illumination, non-uniform intensity, and complex texture noise. An integrated system for the automatic extraction of pavement cracks based on progressive curvilinear structure filtering and optimized segmentation techniques is proposed in this paper. Considering phase congruency and path morphological transformation, a phase congruency guided multi-scale path anisotropy filtering (PCmPA) method is first developed to generate a crack saliency map, significantly enhancing crack structures and eliminating isotropic texture noise. Phase congruency guided multi-scale free-form anisotropic filter (PCmFFA) is then presented as an extended curvilinear structure filter considering context information to enhance PCmPA. Finally, to accurately identify crack pixels and background, the two independent global filtering responses are incorporated with the phase congruency map and integrated into the graph-cuts based global optimization model with an adaptive regularization parameter. Experiments are conducted on two public pavement datasets and a self-captured laser-scanned pavement dataset, with results demonstrating that the proposed method can achieve superior performance compared to six existing algorithms.