Desafios na utilização de técnicas de renderização não fotorrealistas com transferência de estilo com características do cartoon: uma revisão sistemática da literatura

Abstract

Non-Photorealistic Rendering (NPR) or stylized rendering is a technique that has been developed over the years, which, in contrast to realistic image synthesis, aims to generate images or videos of 3D scenes with specific features that convey the idea of a hand-drawn illustration. This application is expanding, common to the fields of computer graphics and image processing, with the purpose of clarifying the principles and concepts of various methods that exhibit differences in style and visual appearance. However, despite the advancements, there is a scarcity of works that present the state of the art on the proposed topic. Therefore, this work aims to conduct a systematic literature review on non-photorealistic rendering techniques that utilize generative adversarial networks for style transfer with cartoon characteristics. We acknowledge that these techniques are constantly evolving, so identifying works that address this topic can be useful for researchers developing new solutions. In the development of this work, a methodological planning was established, resulting in the analysis of 16 relevant articles. These articles were examined, thus allowing for the answering of the established research questions. The obtained results indicate that these techniques are becoming increasingly promising for the field, however, there are still challenges to be overcome, such as the accurate capture of desired characteristics, detail preservation, precise control of the desired level of stylization, and computational efficiency for real-time renderings. Based on the obtained results, researchers can direct their efforts towards overcoming the mentioned challenges and developing new, more efficient and effective solutions.