Enhanced CCA and its Applications in Feature Fusion of Face Recognition

Based on the theory of classical canonical correlation analysis(CCA),by defining class correlation,an enhanced canonical correlation analysis(ECCA) is proposed.If a pattern space has two observation spaces(For any pattern,there are two observation vectors of different kinds),ECCA can find the relevant subspaces of the observation spaces,in which the projections of original random vectors are irrelevant.Experiments demonstrate the superior performance of ECCA fusion over feature fusion algorithm such as CCA,GCCA.