Evolutionary Algorithms For Transferring Properties Between Images Part I: Grayscale Image Colorization

Abstract

In this paper, an evolutionary algorithm (EA) for “colorizing” grayscale images is introduced by evolving color patch transfer process between a source colored image and a target grayscale image. As the general problem of inverting a gray palette to a color palette is a severely under-constrained, ambiguous problem and has no exact, objective solution, human labor and costly semantic knowledge are required. The presented EA attempts to minimize the amount of human work by automatically choosing colored patches from the source image and applying their colors to the grayscale patches of the target image. Furthermore, the best patch matching over all EA parent individuals are recombined in a single multi-sexual recombination scheme to form a single offspring individual. Mutation, on the other hand, forms all other EA individuals. The simple technique of the proposed EA can be successfully and efficiently applied to a variety of images.