Frontal face recognition from video via rank-aware multiple measurement vector recovery

The Sparse Classification approach for image based face recognition assumes that each test sample can be expressed as a linear combination of training samples of the correct class. We propose a video based face recognition approach based on the same assumption. Our formulation requires solving a low-rank row-sparse Multiple Measurement Vector (MMV) recovery problem. Such a row-sparse MMV matrix is low rank as well. Since low rank row sparse MMV recovery is not a well-studied problem, we propose a novel algorithm to solve such it. The experimental evaluation is carried on the VidTIMIT database. The proposed method yields better results than state-of-the-art methods in video based frontal face recognition.