Avenues for the use of cellular automata in image segmentation

Abstract

The majority of Cellular Automata (CA) described in the literature are binary or three-state. While several abstractions are possible to generalise to more than three states, only a negligible number of multi-state CA rules exist with concrete practical applications. This paper proposes a generic rule for multi-state CA. The rule allows for any number of states, and allows for the states are semantically related. The rule is illustrated on the concrete example of image segmentation, where the CA agents are pixels in an image, and their states are the pixels’ greyscale values. We investigate in detail the proposed rule and some of its variations, and we also compare its effectiveness against its closest relative, the existing Greenberg-Hastings automaton. We apply the proposed methods to both synthetic and real-world images, evaluating the results with a variety of measures. The experimental results demonstrate that our proposed method can segment images accurately and effectively.