Efficient location, imaging and recognition of faces by single-pixel camera

Face recognition is a problem with many practical applications. Modern image analysis methods such as object tracking, feature extraction, classification or verification that explore advanced techniques of machine learning and compressive sensing have been used for this purpose. Many of these methods, which are usually applied in image post-processing, are adaptable to fast intelligent viewing with a single-pixel camera. We study such a camera in the context of face recognition with limited information. We seek the optimal basis of patterns for the camera as well as to resolve practical issues related to face localisation and alignment. We compare the use of the Hadamard and eigenface bases for imaging and verification of faces. For the latter task we develop a simple algorithm based on compressive sensing.