Research Highlights

Comic is a world-wide popular art form enjoyed by people of all ages. However, creating a comic book is time-consuming, especially the screening process. There still lack automatic tools to release the artist from this tedious process. So, an automatic comic screening system is highly desired in the comic industry for saving time and labor. Recently, deep learning technologies have greatly advanced the development of various image analysis and image synthesis applications. Nevertheless, the existing deep learning methods generally fail to generate screentones with intensive strokes from simple line drawings due to the large information gap between line drawings and screentoned manga.

The proposed solution introduces a tone-aware semantic layer analysis approach that decomposes the complex line-to-screentone transformation into two sequential tasks: line-to-tone prediction and tone-to-screentone generation. By leveraging tone maps as an intermediate representation, the system simplifies the learning process and improves generation quality. Advanced diffusion-based deep learning models are employed, together with an optimized training strategy that enhances high-frequency screentone details. A large-scale supervised dataset comprising line drawings, tone images, and screentoned manga is constructed to support robust model training. Experimental evaluations, including quantitative metrics and user studies, demonstrate that the proposed method significantly outperforms existing approaches, achieving strong user preference and superior visual quality.