Face photo-sketch recognition based on joint dictionary learning

Abstract

Face recognition technology is widely used in law enforcement agencies. Face photo-sketch recognition is one of possible ways to identify suspects. We propose a method using joint dictionary learning for face photo-sketch recognition. Our method bypasses the image synthesis procedure used by previous joint dictionary learning based methods. Compared with other methods such as coupled dictionary learning which projects features from two different modalities into a common space for recognition, our method does not need extra projections, and avoids the expensive optimization of coupled dictionary learning. By using the cosine distance nearest neighbor classifier, our method performs equally well as coupled dictionary learning based method with much less computation. In the experiments on a popular face photo-sketch database, our method achieves recognition rates higher than or comparable to that of the state-of-art methods.