This paper investigates the use of a simple genetic algorithm, with a brightness and an edge fitness function, to control the motion of objective function cells across an image, which contains geometrically simple objects (squares). The chromosomes are fixed length bit strings code the probability of choice for the direction of motion of a cell. The directions are coded as the amount of rotation the cell performs during each generation. Reproduction is by population substitution using only a crossover operation. Some two hundred cells evolve for approximately one hundred generations. The cells migrate to the edge and/or bright pixels in less than ten generations. The final positions of the cells and their corresponding bit string values are recorded. A fitness sharing function is used to distribute the cells over the objects in the image, in proportion to a particular object's grey level intensity. Hence, convergence and exploitation are avoided and, thus, the maximum amount of exploration of the image is achieved.
展开▼
