A genetic algorithm(GA)-based method for the combinatorial optimization in contour formation

Abstract

Object recognition methods that are based on geometric features are intrinsic to solving the visual pattern recognition problem, where contour feature is one of the most important geometric clues. The biological visual cortex can collect fragmentary edge data, which can be combined into longer, more integrated edges. This is a typical combinatorial optimization problem. Because genetic algorithm (GA) is suitable to solve such problems, it is here used to integrate short line segments into long contour lines by using a graph-based genetic representation and improved genetic operations. The results of the present experiments show that the proposed method can significantly increase the effectiveness of forming long contour lines, which strongly facilitate the realization of the recognition invariance. Longer line contour representation contributes significantly to the formation of the structured semantics of objects, explicit knowledge representation of object recognition, and realization of top-down processing.