AN EXPERIMENTAL-STUDY OF AN OBJECT RECOGNITION SYSTEM THAT LEARNS

The main objective of this paper is to study the performance of a Know ledge-based Object REcognition system that Learns (KOREL). Other objec tives of this paper include, firstly, identifying information about ed ge-junction objects that is particularly useful in recognizing objects in two-dimensional (2D) images. An edge-junction object is an object that can be recognized by the arrangement of its junctions and by whet her each edge is curved or straight. Secondly, this paper presents met hods to acquire (learn) this information automatically and to use it i n object recognition. Thirdly, whereas most previous systems aimed at identifying one or a few target objects, the research reported here ad dresses a more challenging problem, namely, recognizing any edge-junct ion objects known to the system. The number of objects to be recognize d might be large, their appearances in the images might be slightly di fferent from those previously encountered, and they might be partially occluded. KOREL represents characteristic views of a three-dimensiona l (3D) object by models derived from 2D images. Three-dimensional obje cts in a scene are then recognized by matching a 2D image of the scene against object models. KOREL recognizes an object primarily by abstra cting its structure, with representations of less comprehensive struct ures indexing representations of more comprehensive structures. Exact matching is not required, so occlusion and imperfect data are accommod ated.