Automatic Face Caricatures Synthesis and Exaggeration

Face caricatures are widely used in political cartoons and generating caricatures from images has become a popular research topic recently. The main challenge lies in achieving nice artistic effect and capturing face characteristics by exaggerating the most featured parts while keeping the resemblance to the original image. In this paper, a sketch-based face caricature synthesis framework is proposed to generate and exaggerate the face caricature from a single near-frontal picture. We first present an effective and robust face component rendering method using Adaptive Thresholding to eliminate the influence of illumination by separating face components into layers. Then, we propose an automatic exaggeration method, in which face component features are trained using Support Vector Machine (SVM) and then amplified using image processing techniques to make the caricature more hilarious and thus more impressive. After that, a hair rendering method is presented, which synthesizes hair in the same caricature style using edge-detection techniques. Practical results show that the synthesized face caricatures are of great artistic effect and well characterized, and our method is robust and efficient even under unfavorable lighting conditions.