Example-based automatic portraiture

We present a new approach for automatically generating a life-like portrait from a frontal face image. We learn the portraiture from a set of artwork examples. Different from previous texture synthesis and image synthesis works that assumed modeling is homogeneous, Inhomogeneous Markov Random Field Model is employed as the statistical model, and a non-parametric sampling scheme is used to capture the complex statistical characteristics of face image and corresponding artist drawing in this paper. In our strategy, only those pixels corresponding to a portrait point are sampled. Such a strategy is crucial for maintaining facial structure and guaranteeing coherence of portrait lines. Experimental results demonstrate the effectiveness and life-likeness of our approach.