Joint space learning for video-based face recognition

Popularity of surveillance and mobile cameras provides great opportunities to video-based face recognition (VFR) in less-controlled conditions. This paper proposes a joint space learning method to simultaneously identify the most representative samples and discriminative features from facial videos for reliable face recognition. Specifically, we use a mixture modal by learning multiple feature spaces to capture the data variations where the representative samples in each subspace are learned. Actually, this procedure is a chick to egg problem and an alternate algorithm is developed to monotonically optimize the joint task. In addition, randomized techniques are applied to kernel approximations for capturing the nonlinear structure in data, so that both accuracy and efficiency of our method can be improved. The proposed method performs better than the state-of-the-art video based face recognition methods on Honda, Mobo and YouTube Celebrities databases.