Real-time image-based Chinese ink painting rendering

摘要

Chinese ink painting, also known as ink and wash painting, is a technically demanding art form. Creating Chinese ink paintings usually requires great skill, concentration, and years of training. This paper presents a novel real-time, automatic framework to convert images into Chinese ink painting style. Given an input image, we first construct its saliency map which captures the visual contents in perceptually salient regions. Next, the image is abstracted and its salient edges are calculated with the help of the saliency map. Then, the abstracted image is diffused by a non-physical ink diffusion process. After that, we combine the diffused image and the salient edges to obtain a composition image. Finally, the composition image is decolorized and texture advected to synthesize the resulting image with Chinese ink painting style. The whole pipeline is implemented on the GPU, enabling a realtime performance. We also propose some optional steps (foreground segmentation and image inversion) to improve the rendering quality. Experimental results show that our model is two to three orders of magnitude faster, while producing results comparable the ones obtained with the current image-based Chinese ink painting rendering method.