A novel model for face recognition

This paper propose a novel face recognition model which includes three parts. Firstly, Principal Component Analysis (PCA) is adopted to perform dimensionality reduction and decorrelation of face images which enables us to acquire decomposition coefficient with acceptable time in the subsequently stage, namely sparse representation-based classification (SRC). SRC is typically used to represent signal sparsely based on overcomplete dictionary established by base elements which describe certain architectural feature of original signal. To represent face images sparsely and efficiently, we construct overcomplete dictionary using eigenfaces as atoms in accordance with SRC theory. In fact, SRC module can be regarded as an l₁-Minimization problem, which is typically underdetermined and its solution is not unique. At last we employ Homotopy to compute the expansion coefficients effectively and fastly. Experimental results based on Yale face database show the validity of PCA combined with SRC and Homotopy algorithm in face recognition.