| Abstract | Cellular Automata (CAs) are spatiotemporal dynamic complex systems that can serve as an alternative framework for the modeling of physical and biological systems. Furthermore, they constitute intrinsically parallel models of computation which can break through the von Neumann bottleneck and be efficiently realized with special-purpose cellular automata machines. In this study, image processing is the domain of interest throughout: the basic objective is to determine techniques for using CAs to model digital images, and to extract edge features on binary and 256 gray-scale images by utilizing general-purpose hardware. In addition, the rule space, state space, and computation model of the pr0posed CA are investigated in depth. The computation complexity, computation times, and edge-maps quality for the proposed CA—based model are realized. In these respects, a number of well-known edge operators are evaluated and compared. The origins, topology, and modeling categories of the CA5, which are necessary for an understanding of the basic ingredients our work, are investigated. A two-dimensional CA with a regular configuration using a von Neumann neighborhood-based rule is presented, which carries out edge detection on binary images. A transition matrix algebraically defined a transition function associated with the proposed CA is given. It determines the characteristic polynomial, which explains the properties of the CA state space. In addition, two state-vector transition property and an activity relation between Hamming distance and evolution time steps are also comprehensively analyzed to explain the proposed CA state space. Analyses of the CA computation model using a state graph and a finite state machine are provided in depth. Furthermore, a deterministic finite automaton (DFA) associated with the state graph as a recogniser machine is constructed for determining if a particular configuration can be made possible by the given CA rule. The CA rule for edge detection on binary images is extended to carry out 256 gray-scale images. Surprisingly, both are the same, but deal with different state spaces. In addition, the computation complexity of the CA-based model using the analyzing method is investigated, and the results are confirmed through simulation. For evaluating performance characteristics of the proposed CA-based model, a number of distinguished edge detectors (Canny, Laplace, Sobel, and Kirsch) is evaluated and compared in term of computation times, computation complexity, and edge-maps quality. On the basis of key issues in the comparative evaluations, our work shows that the CA-based model provides an optimum edge map on binary images, and on average is better than the compared edge operators for 256 gray-scale images. |