ABSTRACT
Let us give a state-of-the-art example to explain what it means for a segmentation algorithm to be fully automatic and consistent. We consider only a single image here shown in Figure 2.1a, but our analysis is not restricted to single images only. Most segmentation algorithms amount to statistical
FIGURE 2.1 (See color insert following p. 94.) Segmentation results by Jianbo Shi and Jitendra Malik (2000) for the image (a) for the two cases when a number of regions are chosen to be 10 and 60, as shown in (b) and (c), respectively. If the trees are the object of interest, the segmentation in (b) is suitable, whereas if the horse is of interest, then the segmentation given in (c) is more suitable.
