A genetic aproach for automated image generation: grayscale image generation

Abstract EN

Non photorealistic rendering is a new research field in the areas of computer graphics.

The goal is to give a more natural feel to computer generated images, by simulating various artistic techniques and to give the sense of an image without reproducing it.

In this paper, we present a new evolutionary approach to non-photorealistic rendering of 2D black/white and grayscale images.

The goal is to generate a painting that is close to a given input images.

This problem can be formalized as a high-dimensional optimization problem, with local minima.

We have developed a genetic algorithm that modifies the traditional uniform crossover to spread out vital genes at the expense of lethal genes rather than exchanging them between matting parents.

A vital or lethal gene can be determined via a threshold field associated with each pixel gene that indicates the distance between a chromosome gene and the corresponding input image pixel.

The proposed evolutionary painting framework demonstrates good results and achieves reasonable convergence.