A Markov random field for rectilinear structure extraction in pavement distress image analysis

This paper deals with the detection and extraction of poorly contrasted rectilinear structures in textured areas, using a Markov random field model. The application is in the analysis of pavement distress, and more particularly pavement cracks. A local crack detection is first performed, where the pavement texture is seen as additive correlated noise. The resulting line image is then projected onto a regular lattice composed of straight line segments. A graph structure is associated with this lattice, which allows the definition of a Markovian crack model, where sites are no longer the image pixels, but straight line segments. The model is used to determine the location and shape of the rectilinear structures, with a given orientation, in the observed lattice. The actual defects can then be extracted by simple post-processing.