SYNTACTIC AND SEMANTIC METHOD FOR IMAGE SEGMENTATION
Abstract
The problem of image segmentation is studied. A syntactic approach is proposed. Edge, line (curve), corners are identified as special type of image patterns. A set of ideal edges, lines and corners are used to infer a tree grammar. Minimum-weighted distance SPECTA algorithm is used to find a best match between ideal patterns and actual patterns in the image. A region detector has also been developed as an alternate approach to the problem of image segmentation. This is a sementic approach. Regions are approximated by linear and quadratic surfaces. Attributes, which reflect the properties of a pixel belonging to a region, are assigned to pixels. Concept of sementic distance is introduced. Finally, the problem of boundary detection in texture is considered. Again, a syntactic approach is proposed. Two or more textures give rise to distinct patterns at the border where they meet. Tree grammar or stochastic tree grammar can be used to generate these patterns. Minimum-distance maximum-likelihood SPECTA is used to detect these patterns. The performance of these approaches to the problem of image segmentation is illustrated by means of experimental results obtained with real-world images.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.