Medical image segmentation is the crucial process for three-dimensional (3D) modeling. Many studies have claimed to suggest more accurate ways of segmentation. However, such research seems to have the drawback that it works against the parsimony principle. These methods have to resegment the regions of interest (ROIs) for the already segmented data, without taking advantage of the vast amount of past segmentation results. The present report proposes a "segmentation by evolution" method. It is an advanced segmentation paradigm that incorporates the existing segmentation results. The "segmentation by evolution" paradigm was applied to the color images of the Visible Human data.