SHAPE REPRESENTATION AND RECOGNITION WITH INCOMPLETE INFORMATION
Abstract
Many shape recognition methods require complete knowledge about all possible known patterns and about the unknown pattern. In many applications, complete specification of both the known and unknown shapes may not be possible or desirable. In some applications, complete views of the unknown objects may not be available. Classification must be performed using the incomplete information provided by the identifiable portions of the object. Another situation involves the identification of an unknown object as belonging to a class of objects, rather than as a specific object. In this case, the detailed structure of the individual class elements is not desired, so the classification is made using information which represents the basic structure of the object. A third case arises when objects are represented by range data. A full three dimensional model of the object is known, but the range data is available only for the surfaces detected by the sensor. Classification must be achieved using the incomplete three dimensional information provided by these surfaces. Each of these three examples requires shape recognition with incomplete information, and a shape representation into which the incomplete information can be incorporated. These three areas of shape representation and recognition are investigated. Shape representations are examined and recognition techniques developed and evaluated. Specifically, a partial shape recognition algorithm using dynamic programming has been developed. The algorithm has several shortcomings which are discussed, and refinements and extensions of the algorithm to address these shortcomings are presented. The concept of identifying an object as a member of a class is examined in the context of aircraft recognition. The objective is to decide if an object is an aircraft, regardless of aircraft type, and to estimate the orientation. A representation and recognition technique based on skeletons is presented, as is a study of the orientation estimate accuracy. A representation for three-dimensional objects using spherical harmonics is presented and extended to the representation of objects described by depth data. This technique is used to represent and recognize several simple three dimensional objects which are described by simulated depth data. Experimental results are discussed and evaluated.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.