Automatic Construction of Image Transformations to Produce Variously Stylized Painterly Images
Keita Nakayama, Tomoharu Nagao · 2013
We describe a new method that generates images that represent various painting styles by constructing the process of painterly rendering, using ACTIT (Automatic Construction of Tree-structural Image Transformation). ACTIT is an image processing system that automatically constructs image transformations by using Genetic Programming (GP) and image processing examples. The constructed image transformations can then be applied to other images. In our method, we add two extensions to ACTIT so that it can be applied for constructing the painterly rendering process. The first extension comprises image processing filters that append non-photo realistic effects. The second is our proposed new fitness function of ACTIT for evaluating the features of the images used for painterly rendering. The results of experiments in which we used three painting style examples show that our method constructs three image transformations that produce the respective painting styles.