On the use of colour-based segmentation in evolutionary image composition

Abstract

Evolutionary algorithms have been widely used in the area of creativity in order to help create art and music. We consider the recently introduced evolutionary image composition approach based on feature covariance matrices [1] which allows composing two images into a new one based on their feature characteristics. When using evolutionary image composition it is important to obtain a good weighting of interesting regions of the two images. We use colour-based segmentation based on K-Means clustering to come up with such a weighting of the images. Our results show that this preserves the chosen colour regions of the images and leads to composed images that preserve colours better than the previous approach based on saliency masks [1]. Furthermore, we evaluate our composed images in terms of aesthetic feature and show that our approach based on colour-based segmentation leads to higher feature values for most of the investigated features.