Representation

Abstract

The role of the initial segmentation is to partition a range image into smooth surface regions and to trace the contours of surface discontinuity. After segmenting the image into smooth surface regions and surface discontinuity contours, the next step is to find an appropriate representation for the segmented image. The representation should be such that it aids higher level cognitive visual functions. A representation at this level is termed a feature-level representation and the process of arriving at such a representation is referred to as feature extraction. In order to recognize objects in the scene, it is essential for the computer vision system to have a pre-stored representation of the object(s) it expects to encounter in the scene. The internal representation of an object in the memory of the vision system is referred to as an object model, and the process of formulating and implementing an object model is referred to as representation modeling. For the purpose of recognition, there should be a one-to-one correspondence between the feature-level representation and the object model representation. In this chapter the term representation will be used to denote both the feature level representation and the object model representation. Since objects encountered in range images are primarily three-dimensional with smooth C² surfaces, geometric representation schemes for three-dimensional objects are of particular interest.