Semi-supervised Learning of Caricature Pattern from Manifold Regularization

摘要: Automatic caricature synthesis is to transform the input face an exaggerated one. It becoming interesting research topic, but it remains open issue specify caricature's pattern for face. This paper proposed a novel prediction method based on MR (manifold regularization), which comprises three steps. Firstly, we learn by manifold dimension reduction, and select some low dimensional as labels corresponsive true faces. Secondly, regularization performed build semi-supervised regression between faces labels. In third step of offline phase, mapped label learnt regressive model, further transformed parameters locally linear reconstruction algorithm. approach takes advantage structure lying in both caricatures. Experiments show that, represents well model from can predict target successfully.