Unsupervised segmentation of road images

This paper presents a region-based segmentation algorithm which can be applied to various problems since it does not require a priori knowledge concerning the kind of processed images. This algorithm is based on a split and merge method. The splitting algorithm uses a homogeneity criterion based only on grey levels. We modeled exploited fields by Markov Random Fields (MRF), the segmentation is then optimally determined using the Iterated Conditional Modes (ICM). Input data of the merging step are regions obtained by the splitting step and their corresponding features vector. The originality of this algorithm is that texture coefficients are directly computed from these regions. Thus, a region-based segmentation algorithm using texture and grey level is obtained. Results from road scenes without white lines are presented.