Multi-View Representation Based Face Sketch Synthesis

Abstract

Face photos and sketches are in two different modalities and synthesizing face sketches from photos plays an important role in law enforcement and digital entertainment. Although many face sketch synthesis methods have been proposed in recent years, most of these methods choose neighbor image patches based on image intensities. However, it is not appropriate to represent the similarity of two image patches merely according to image intensities, especially under uncontrolled environments such as varying illumination conditions. In this paper, we propose a multi-view representation of image patches for face sketch synthesis. It first constructs a high dimensional multi-view feature vector through a hierarchical framework including multiple filters and multiple local features. Then an unsupervised dimensionality reduction method is used to reduce the cost of computation and model storage. Finally traditional face sketch synthesis methods can be applied based on the proposed representation. Experimental results on the Chinese University of Hong Kong (CUHK) face sketch database and celebrity photos from the Internet illustrate that the proposed strategy improves the performance of the state-of-the-arts.