Image-Guided Rendering with an Evolutionary Algorithm Based on Cloud Model.

Tao Wu

doi:10.1155/2018/4518265

Abstract

The process of creating nonphotorealistic rendering images and animations can be enjoyable if a useful method is involved. We use an evolutionary algorithm to generate painterly styles of images. Given an input image as the reference target, a cloud model-based evolutionary algorithm that will rerender the target image with nonphotorealistic effects is evolved. The resulting animations have an interesting characteristic in which the target slowly emerges from a set of strokes. A number of experiments are performed, as well as visual comparisons, quantitative comparisons, and user studies. The average scores in normalized feature similarity of standard pixel-wise peak signal-to-noise ratio, mean structural similarity, feature similarity, and gradient similarity based metric are 0.486, 0.628, 0.579, and 0.640, respectively. The average scores in normalized aesthetic measures of Benford's law, fractal dimension, global contrast factor, and Shannon's entropy are 0.630, 0.397, 0.418, and 0.708, respectively. Compared with those of similar method, the average score of the proposed method, except peak signal-to-noise ratio, is higher by approximately 10%. The results suggest that the proposed method can generate appealing images and animations with different styles by choosing different strokes, and it would inspire graphic designers who may be interested in computer-based evolutionary art.

Highlights

Evolutionary design is an area in which artificial evolution is applied to problems in engineering, design, and the fine arts
We propose an image-guided rendering with an evolutionary algorithm based on cloud model, and our intentions are two-fold: (1) How can the image-guided rendering be produced with a cloud modelbased evolutionary algorithm? (2) How different types of strokes can affect the aesthetic qualities of the rendered images?
The proposed framework can generate interesting images and animations with the nonphotorealistic rendering (NPR) style based on a given input image

Summary

Introduction

Evolutionary design is an area in which artificial evolution is applied to problems in engineering, design, and the fine arts. One active research topic is the generation of images using evolutionary algorithms [1,2,3,4,5,6,7]. It would be beneficial to researchers to develop and test the evolutionary algorithms for creating interesting and aesthetical images and videos. A multiagent based art production framework is proposed, which generates images using multiagents with chaotic dynamics features [14]. A methodology is applied for evolving image variants to maximize diversity in image feature metrics [15], and a cooperative coevolution strategy based on the Parisian evolution approach is introduced to produce artistic visual effects from an input image [16]. The problem of evolving images without human-in-the-loop is still an open issue in evolutionary art; sustained effort is required for autonomous evolutionary art

Methods

Results

Conclusion