An art-oriented pixelation method for cartoon images

Peng Lei, Shuchang Xu, Sanyuan Zhang

Abstract: Pixel art has evolved from a primitive computer image presentation to an independent digital art style. It is widely used on the internet, for graphic user interface (GUI) design, and game industries. Existing pixelation tools and algorithms generate pixel images with artifacts, color clutter, blurring, and a lack of aesthetics. Generally, aesthetics are the dominant concern for pixel art. In this paper, an art-oriented pixelation (AOP) algorithm is proposed to effectively retain the main features of the original image content and the integrity of essential details with the artistic and aesthetic styles. At the same time, the AOP algorithm enables high-quality pixel image generation of arbitrary size without paired datasets and model training effort. The experimental results demonstrate that the pixel image generated by the AOP algorithm outperforms existing algorithms and tools in terms of aesthetics.