Shape description and boundary restoration.
Item
-
Title
-
Shape description and boundary restoration.
-
Identifier
-
AAI9009744
-
identifier
-
9009744
-
Creator
-
Jankowski, Mariusz.
-
Contributor
-
Adviser: George Eichmann
-
Date
-
1989
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Engineering, Electronics and Electrical | Physics, Electricity and Magnetism
-
Abstract
-
There is a growing interest in robotics, machine vision and computer graphics fields in developing efficient representations of three-dimensional (3D) objects. A choice of a representation scheme is critical to the efficient manipulation and storage of image data. Of special interest are geometrical representations. Since surfaces are what is seen, surface representations are important to computer vision. The enclosing surface, or boundary, of a well behaved 3D rigid solid unambiguously specifies the object. Most importantly, the boundary captures the notion of shape, which is an intrinsic property of solid objects. For purposes of pattern recognition, it is of great interest therefore, to represent shape numerically. The resulting so-called shape descriptors provide a canonical, viewer independent image representation that greatly simplifies the recognition task. The notion of similarity of objects, based on their shape characteristics, is simplified to a measurement of distance in an N-dimensional feature space. A new three-dimensional (3D) shape descriptor is calculated from the differential characteristics of a solid's bounding surface. Classification results for polyhedral and cylindrical solids are presented.;A common and mainly unsolved problem in image processing is occlusion. Error-correction that restores the boundary to its original shape makes shape description possible under conditions of partial boundary visibility. An error-correcting associative memory is used to solve the restoration problem. The associative memory is a special case of a neural network (NN). The generalizing and associative properties of NNs are especially well-suited to handle problems which involve distorted, "blurry", and unpredictable data like the boundary reconstruction problem. Examples are given of boundary restoration for three-dimensional (3D) range data images of solids and two-dimensional (2D) boundary curves. Both, a Kohonen associative memory and a back-propagation NN are used.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
