Locating image background by monotonic tree

In this paper we discuss an approach to locating image background, which is based on monotonic tree, a derivation of contour tree for discrete data. We first make two assumptions about image background. As compared with the foreground shapes, the background regions area (1) smoother in intensity; and (2) peripheral in location. In an input image, the harshness of the foreground is characterized by intensity peaks and valleys, which are modeled in monotonic tree as small textons whose altitudes are high. Based on this, we can find the texton pattern for foreground objects. Then the foreground region, and thus the background region, are rendered. Experiments. show that our scheme can effectively locate background in images which. satisfy the two assumptions above.