| Efficient face recognition system is one which delivers high accurate results to the end user in good time. For this purpose, face representation must be robust, discriminative and also of low computational cost in both terms of time and storage requirement. Motivated by the success of the recently proposed feature descriptor called Patterns of Oriented Edge Magnitudes (POEM), which balances three concerns, this paper aims at enhancing its performance with respect to all these criteria via learning process. For every pixel, the POEM features are built by applying a self-similarity based operator on oriented magnitudes, calculated by accumulating a local histogram of gradient orientations over all pixels of image cells, centered on the considered pixel. In this work, we first optimize the parameters of POEM and then apply the PCA dimensionality reduction technique followed by a Whitening transform to get a more compact, robust and discriminative descriptor. The very strong results for face recognition, on both constrained (FERET) and unconstrained (LFW) datasets prove the efficiency of our algorithm. Impressively, our algorithm is at least 27 times faster than those based upon Gabor filters. |
Récupérer au format
