PT-ACTIT; Parameter Tunable-Automatic Construction of Tree-structural Image Transformation
Wataru Fujishima, Tomoharu Nagao · The Journal of The Institute of Image Information and Television Engineers · 2005
In image processing, selecting image processing filters to satisfy the conversion from an original image to its target image is difficult. We previously developed an image processing system, ACTIT, that constructs image processing filters to satisfy the given conversion from original to target image. ACTIT uses GP, genetic programming, and it constructs tree-structured image processing filters with pre-defined well-known filters including several, which require numerical parameters. However, optimizing both structural and parameters at the same time is difficult for conventional GP. Thus, we developed PT-ACTIT, which can optimize structural and numerical parameters simultaneously. GA, genetic algorithm, was used to optimize nuemrical parameters. Experimental results using PT-ACTIT show that we can construct a new filter that cannot be constructed using only pre-defined filters.