A PARALLEL APPROACH TO SYNTACTIC IMAGE ANALYSIS (PATTERN RECOGNITION, RELAXATION PROCESS, TREE GRAMMAR, TRANSLATION)
Abstract
Syntactic methods and relaxation processes are two important techniques for image analysis. Both of these techniques possess the capability of handling the structural information contained in a pattern or an image. However, the syntactic methods are basically sequential processes and the relaxation processes are basically parallel processes in nature. In this research, it is attempted to bridged these two techniques, i.e., applying techniques of the relaxation processes to the syntactic methods, so that merits of the parallel nature of the relaxation processes can be incorporated with the rigorousness of the syntactic methods. For nonstochastic cases, the discrete relaxation processes can be used to interpret the segmentation and parsing process of the context-free languages. For stochastic cases, the concept of matched filters is proposed. A matched filter is designed to match a given grammar, which can enhance those patterns generated by the grammar and suppress the other patterns. Using stochastic tree grammar as the context-generation model, the compatibility coefficients between neighboring primitives can be derived when the context-generation process is in the equilibrium state. An upper bound for the number of iterations required by the matched filter is also found. Using the language translation theory, the concept of matched filter is also extended to the time-varying image sequences analysis. Matched filters operating in either space domain or time domain can be designed. Relationships between these two filters are also studied.
Degree
Ph.D.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.