Color image segmentation with watershed on color histogram and Markov random fields

Watershed algorithm is traditionally applied on image domain. It fails to capture the global color distribution information, In this paper, a two-step segmentation framework is proposed. Watershed on color histogram is the first step that is aimed at solving the problem of improper color clustering caused by color clusters with irregular shapes. L*a*b* color space is adopted in this step because it is more consistent with human perception. After the coarse segmentation result obtained via watershed, the highest confidence first (HCF) algorithm for Markov random fields is taken as the second step to refine the coarse result to get continuous regions. Experiments with real color images show that the proposed two-step segmentation framework is efficient.