Skip to Main Content
A new method for color image segmentation is presented. It combines mean shift with watershed algorithm to get robust results. First, mean shift procedure is used to find the highest density regions which correspond to clusters centered on the modes (local maxima) of the underlying probability distribution in the feature space. The principal component of each significant color is extracted by mode. Second, homogeneous regions corresponding to the modes are as markers to label an image, then marker-controlled watershed transformation is applied to the image segmentation. The segmentation of blood cells is discussed. The input parameters are a few initial color markers that represent significant colors. By combining both methods, the oversegmentation is able to be prevented, touching and overlapping nucleated cells can be separated, and running time is reduced too. The proposed algorithm is very robust to different color space, varied preparation and illumination for blood cell images. It is suitable to segment color microscope images.
Machine Learning and Cybernetics, 2003 International Conference on (Volume:5 )
Date of Conference: 2-5 Nov. 2003