Improvements on CCA Model with Application to Face Recognition

Two new methods for combination feature extraction are proposed in this paper. The methods are based on the framework of CCA in image recognition by improving the correlation criterion functions. Comparing with CCA methods, which can solve the classification of high-dimensional small size samples directly, being independent of the total scatter matrix singularity of the training simples, and the algorithms' complexity can be lowered. We prove that the essence of two improved criterion functions is partial least squares analysis (PLS) and multivariate linear regression (MLR). Experimental results based on ORL standard face database show that the algorithms are efficient and robust.