Segmentation by morphological reconstruction and non-linear diffusion

Existing methods for color image segmentation using diffusion can't preserve contour information and noises with high gradients become more salient as the number of times of the diffusion increases, resulting in If over-segmentation when applied to watershed. This paper proposes a new method for color image segmentation by applying morphological operations together with nonlinear diffusion. Morphological reconstruction works very effectively for the removal of high frequency noises, while nonlinear diffusion can remove low frequency noises. Based on the idea, the input image can be transformed into a simplified image which preserves contour information with noises removed. With gradients computed from this simplified image, watershed algorithm is applied to get the segmentation. Experiments show that color images are segmented very effectively without over-segmentation.