Abstract
A multi-agent framework inspired by natural and physical systems is presented for data discovery and image enhancement. The input image is represented as a topographic landscape upon which a large population of independent simple, reactive, mobile agents resides. The local landscape configuration presents stimuli to each agent, influencing agent behaviour and resulting in changes in agent orientation and movement. Individual agents deposit a trail as they move and leave specific marks in response to stimuli above a certain threshold. The parallel interactions of the agent population and the image landscape result in emergent patterns of trails and marks being generated, corresponding to global (population level) perception of the original image. The emergent patterns exhibit image feature extraction and represent an indirect processing of the input image. External environmental pressures may be applied to the emergent patterns to further amplify the feature extraction. The framework represents a decentralised approach to image enhancement which is extensible. Different types of agent may be developed to perform different image processing functions, or other problems whose definition and solution may be represented as spatial patterns. Results including binary image processing, greyscale enhancement, colour image processing and related spatial processing problems are presented.