Dual space based face recognition using feature fusion
Arpita Patra, Sukhendu Das · 2006
The authors propose a new face recognition technique by combining information from null space and range space of within-class scatter of a face space. The combination of information at feature level poses a problem of optimally merging two eigenmodels obtained separately from null space and range space. The authors use two different methods: covariance sum and Gramm-Schmidt orthonormalization to construct a new combined space, named as dual space, by merging two different set of discriminatory directions obtained separately from null space and range space. The authors employ forward and backward selection techniques to select the best set of discriminative features from dual space and use them for face recognition. Experimental results on three public databases, Yale, ORL and PIE will show the superiority of our method over a face recognition technique called discriminative common vectors (DCV) (Cevikalp et al., 2005), based only on the null space of within-class scatter.