Dual-feature Bayesian MAP classification: Exploiting temporal information for video-based face recognition

Abstract

Machine recognition of faces in video is an emerging problem. Following recent advances, conventional exemplar-based schemes and image set approaches inadequately exploit temporal information in video sequences for the classification task. In this work, we propose a new dual-feature Bayesian maximum-a-posteriori (MAP) classification method for face recognition in video sequences. Both cluster and exemplar features are extracted and unified under a compact probabilistic framework. To realize a non-parametric solution, a joint probability function is modeled using relevant similarity measures for matching these features. Extensive experiments on two public face video datasets demonstrate the good performance of our proposed method. © 2012 Springer-Verlag.