Abstract

Image processing is generally acknowledged as a computationally demanding task. Distributed mobile agents is a form of parallel processing that is gaining attention as a means of increasing the efficiency of many image processing tasks. In addition, a swarm of highly autonomous agents is more reliable in satisfying a task than one sequential process; the failure of a couple of agents is unlikely to stop the rest of the swarm from achieving the overall task. This report discusses a specific application of autonomous agents to the problem of pattern pre-processing. Specifically, we design and build a system that uses a number of behaviour-based agents to detect, segment and thin, as well as describe a hand-written patter (character or word). Each agent has a subsumption type architecture that employs a number of behaviors. These behaviors are activated when certain conditions are satisfied. Every agent has a local coordination unit that arbitrates between the various behaviors to decide which one of the behaviors is active at any one time. The system has a whole a global coordination unit that creates and spreads the various agents over the image, then acts as a conduit of (light) communication between the agents, until the overall task is done, and all last agent dies. In this work we have introduced a number of innovations including a new method of simultaneous thinning and segmentation and a simple but effective development of the classical Fourier descriptor method that works better for open curves. We have tested our system on Arabic characters and Chinese glyphs. The agents were able to find patterns and then work on parallel to extract and segment the centerline in a few seconds. A fixed feature vector of ten descriptors is computed for each segment. The reconstruction process showed that this vector is capable of retrieving the whole shape of the signature.* *This dissertation is a compound document (contains both a paper copy and a CD as part of the dissertation)