Sparse Representation for Face Recognition

Sparse representation has attracted a great deal of attention in the past decade. Famous transforms such as discrete Fourier transform, wavelet transform and singular value decomposition are used to sparsely represent the signals. The aim of these transforms is to reveal certain structures of a signal and representation of these structures in a compact form. Therefore, sparse representation provides high performance in the areas as diverse as image denoising, pattern classification, compression etc. All of these applications are concerned with a compact and high-fidelity representation of signals. In this thesis, we consider the classical face recognition problem. This application is more concerned with the semantic information of image signals. It is shown that a sparse representation based framework is a possible way to tackle this problem. We also propose a new approach for face classification which is based on task driven dictionary learning.