Multiple representation of complex intensity changes for image segmentation

Abstract

In this paper we present a method for finding a multiple representation of complex intensity changes that is useful in image segmentation. In computer vision, it is often required to separate objects from background in images taken under conditions of poor and non uniform illumination. In these situations, it is well-known that many of the intensity changes do not correspond to the object's boundaries but to other factors (mainly shadows and brightness variation) that complicate the segmentation process. With this in mind, we propose a segmentation process that proceeds in stages where one of the most important is the local description of the intensity changes giving rise to a multiple representation where many clues are accentuated. Other higher level stages act to select, manipulate, and combine clues in agreement with certain strategies (decision rules) which can be defined based on previous knowledge or heuristic information in order to segment the entities of interest.