Comparative study of compressive classification-based approaches for face recognition

Feature extraction and classification are two essential components in face recognition. Feature extraction is a process to reduce the original input high-dimensional data and reserve the crucial information. Considering the problem that the human face image is high-dimensional, dimensionality reduction (DR) methods can be employed to obtain low-dimensional data for recognition. Eigenspace-based method is one of the most successful methodologies for DR. Recently, the development of compressed sensing theory leads to a new direction for DR. The aim of this paper is to summarize the classical DR and classification methods. In addition, with respect to DR and classification respectively, two methods with improved recognition rate will be proposed. Experiments are carried out with two popularly utilized face databases (i.e., ORL and Yale) to demonstrate the performance of different methods.