Comparison of Scoring Methods for Interactive Evolutionary Computation based Image Retouching System

Interactive evolutionary computation (IEC) is a useful tool to automate human's creative process such as image drawing, music composition, and so on. In this paper, we apply the IEC to generate a new image from the original one using a set of filters. In professional image editing software, there provides us with a lot of filters and sometimes the sequential use of several filters produces interesting results. Usually, searching for the useful combination of filters is a kind of trial-and-error tasks. Or, only experts can define a good combination of filters from their long experience. In this work, our system incrementally adds or deletes filters based on user's evaluation on the results of image retouching. Recently, people take photos using their smartphones and it is not a trivial to give a number of consecutive evaluations to guide the IEC search. Based on the observation, we try to compare several scoring methods for the IEC to get the best results for the image retouching system. They are a simple binary scoring (Good & Bad), discrete multiple choices (Five Star), and real-valued scoring (Slider). Experimental results on five human subjects show that the Good & Bad is appropriate for the retouching system.