Face recognition via robust face representation and compressive sensing

Abstract

Many face recognition methods devote to the feature selection of training images while pay little efforts on the image representations. This paper proposes a robust face representation using the amplitude projection. The face recognition task can be solved using compressive sensing(CS) theory. First, the amplitude projections capture the horizontal distributions and the vertical distributions of the training faces. We form the projection image based on the horizontal and vertical distributions. Then robust face representation can be described as the input image combined with the projection image. Second, we fit the face recognition task into the compressive sensing framework. Due to the robust face representation and the excellent theory of CS, our approach gives better results when comparing with the state-of-art CS face recognition method. Experiments conducted on two well-known publicly face database verify the accuracy and efficiency of our approach.