Face recognition based on the feature fusion in fractional Fourier domain
Huijing Sun, Enqing Chen, Lin Qi · 2014
Face recognition has become an active research area which has attracted many researchers' attention. In this paper, a new method is proposed, and it selects features in the fractional Fourier domain for face recognition. This new method selects the transform orders by computing the trace ratio of each transform order, then merges three orders' amplitude information of two dimensional discrete fractional Fourier transform (2D-DFrFT) by locality preserving canonical correlation analysis (LPCCA). Multiple orders' amplitude information fusion (MOAF) can not only solve the problem that the single feature cannot represent the face structure adequately, but also can avoid the sensitivity to the nearest neighbor selecting of LPCCA. Experiments comparing the proposed approach with some other popular methods on the AR and CMU-PIE database show that the proposed method consistently outperforms others.