Wavelet decomposition tree selection for palm and face authentication

In this work, we study the usefulness of multi-resolution analysis for the face and palm authentication problems. The images are decomposed into frequency subbands with different levels of decomposition using different wavelets. We adopt as features for the authentication problem, the wavelet coefficients extracted from some “selected” subbands of several wavelet families. We propose to use a multi-matcher where each matcher is trained using a single subband, the matchers are combined using the “Max Rule”. The band selection is performed by running Sequential Forward Floating Selection (SFFS). Moreover, several linear subspace projection techniques have been tested and compared. Experiments carried out on several biometric datasets show that the application of Laplacian EigenMaps (LEM) on a little subset of wavelet subbands (chosen by SFFS) permits to obtain a low Equal Error Rate.