Abstract
Over the past few years, studies have been performed on filter generation methods for image processing with genetic algorithm (GA) or genetic programming (GP). The existing methods generate various filters by combining existing primitive filters. Although filters generated by GP show better output quality than filters generated by GA, chromosome size becomes huge in GP. Uncontrolled chromosome growth, called bloat, makes filter application time worse, and makes it impossible for users to analyze and modify generated filters. This paper proposes a GP based image filter generation method that tunes numeric parameters of the primitive filters and restrains surplus chromosome growth. Experiments with traffic sign extraction problem showed the filters generated by the proposed method had almost the same output quality as filters generated by preceding GP with almost the same size of filters generated by GA.
