Generation of cartoon-style bas-reliefs from photographs

摘要

This paper describes a new algorithm that generates a cartoon-style bas-relief surface from photographs of general scenes. Most previous methods for bas-relief generation have focused on accurate restoration of input 3D models on a background plane. The generation of bas-reliefs with artistic effects has rarely been studied. Considering that non-photorealistic rendering (NPR) techniques are currently very popular and 3D printing technology is developing rapidly, extending NPR techniques to the generation of a bas-relief surface with artistic effects is natural and valuable. Furthermore, cartoon is a basic non-realistic and artistic style familiar to general users. From this motivation, our method focuses on generating a cartoon-style bas-relief surface. We use the lens blur function of Google Camera, which is a smartphone application, to obtain a photograph and its depth map as inputs. Using coherent line drawing and histogram-based quantization methods, we construct a depth map that contains the salient features of given input scenes in abstract form. Displacement mapping from the depth map onto a thin plane generates a cartoon-style bas-relief. Experimental results show that our method generates bas-relief surfaces that contain the characteristics of cartoons, such as coherent border lines and quantized layers.