Genetic sampling in eigenspace for 3D caricature synthesis

Caricature synthesis is becoming one of the most challenging research problems nowadays, especially for 3D caricature. In this paper, we present a novel approach for 3D caricature synthesis adopting genetic sampling in eigenspace. Our approach aims to generate 3D caricature which has not only exaggerated effect but also facial feature of real face. Firstly, we build 3D caricature components eigenspaces by Principal Component Analysis (PCA). Moreover, genetic algorithm (GA) is used to sample optimal projections in the eigenspaces for 3D real face components, and then we reconstruct 3D caricature components by the projections. At last, we refine the 3D caricature using PCA-based sampling approach and Kriging interpolation. The experiment results show that our approach can effectively synthesize interesting and harmonious 3D caricature.