Example-based facial portraiture style learning

An example-based approach for facial portrait style learning is proposed. By learning a training set with the same style, this method can generate new portraitures which are similar to this style. To obtain facial features with high quality, there are two key elements in this paper: Using an Inhomogeneous Markov Random Field Model (MRF) and a nonparametric sampling scheme to learn the statistical relationship between the original images and the corresponding drawings; by identifying the facial contour area, the facial structure is trained from examples independently. Therefore, the output portraits can obtain more details with a clear and complete facial contour, while reducing the noise. Furthermore, an improved multi-samples texture synthesis method is also proposed to speed up the texture synthesis process without loss of the detail. Experimental results show that this approach is more efficient especially in the large image size and can generate satisfying new portraits of the desired styles.