A New Method to Segment Road Image and Produce Road Network

A new strategy is presented for partitioning roads from other objects and producing road networks.The fist step is to find the direction at every pixel,along which the gray value and texture are most consistent.The descriptions of texture in this direction consist of the feature vector.An improved MRF model is then applied to the feature space to accomplish segmentation.The segmented rough road bands are smoothed by an improved line segment matching method,meanwhile,irregular speckles constituted of misclassified pixels are removed.Next,the central lines of roads bands are detected by morphological approaches.Connecting central lines finally produces road networks.Experiments show the efficiency of the proposed method.