A CELLULAR AUTOMATA APPROACH TO COMPUTER VISION AND IMAGE PROCESSING.

摘要

This dissertation describes a system called HEXVIS which performs operations on scenes. All operations are carried out in a hexagonal array of cellular automata-like objects which corresponds to that scene. The system can perform (for instance) the following tasks: recognition of edges, corners, and axes of symmetry, texture discrimination, determination of areas, and generation of Voronoi tessellations.; First, the scene is embedded in the hexagonal array, then, the cells pass messages describing the cells' contents to their neighbors which, in turn, pass them on. As these messages pass through cells, they can interact with each other and with the contents of the cell in which they find themselves. The cells all perform the same operation or group of operations in parallel on their visiting messages. As a consequence of these operations, the states of some cells change in a way which indicates that they correspond to "interesting" parts of the scene.; This process can be repeated recursively using the altered states of the cells as new messages to be broadcast.; In addition a new algorithm is presented which shrinks binary scenes in hexagonal array. It is proved that all scenes with (at most) simply connected holes are transformed into a set of isolated points, each corresponding to a connected object in the original scene. This shrinking algorithm is embedded in HEXVIS.