Latent face model for across-media face recognition

Across-media face recognition refers to recognizing face images from different sources (e.g., face sketch, 3D face model, and low resolution image). In spite of promising processes achieved in face recognition recent years, across-media face recognition is still a challenging problem due to the difficulty of feature matching between different modalities. In this paper, we propose a latent face model that creates mappings from a hidden space to different media space. Images from different media of the same person share the same latent vector in hidden space. A coupled Joint Bayesian model is used to calculate the joint probability of two faces from different media. To verify the effectiveness of our proposed method, extensive experiments conducted on various databases: self-collected low-resolution vs. high-resolution database, sketches vs. photos databases, 3D face model vs. photos on LFW database. Experimental results show that our method boosts the performance of face recognition with images from different sources.