Road Surface Cracks Detection Using Unsupervised Strategies
Road Surface Cracks Detection Using Unsupervised Strategies, Proc Portuguese Conf. on Pattern Recognition - RecPad, Aveiro, Portugal, Vol. ., pp. . - ., October, 2009.
Digital Object Identifier:
This paper describes and compares two unsupervised classification strategies to detect cracks on flexible road pavement surface images. The first strategy uses a Bayesian classifier; the second is based on one-class classifiers. A simple two dimensional feature space is considered, exploiting the mean and the standard deviation of the pixel’s gray levels, computed for non-overlapping image regions. For both strategies a bivariate class-conditional normal density is adopted, for stochastic data modeling, as it produces a good description of the data. Several normalization steps are proposed, to achieve better final results. Experimental crack detection results are presented based on real images taken from Portuguese roads.