Fast face recognition based on KDDA and SRC

Recently, the sparse representation-based classification (SRC) is getting more and more attention in many fields, such as pattern classification, which has been successfully applied to face recognition. However, the essence of SRC, within face recognition, is to use the linear combination in the same level to train and test samples to represent this one. Firstly, the raw data is nonlinear. Using some high resolution images, the sample space will be too big, and "small sample size problem" (SSS) will appear under the high dimension. Secondly, SRC is mainly through an over-completed to obtain the sparse representation of the test sample. On the condition of large data, the computational complexity will seriously affect its performance. To solve these problems, we propose kernel direct discriminant analysis (KDDA), which maps the original nonlinear face subspace into a low-dimensional linear face feature subspace. On this base SRC is performed. Finally, extensive experiments on database are conducted. Experimental results show that our method significantly improves the recognition speed compared with the original SRC, which achieves comparable or even better recognition rates.