A New Approach in Solving Illumination and Facial Expression Problems for Face Recognition

In this paper, a novel dual optimal multiband features (DOMF) method is presented to increase the robustness of face recognition system to illumination and facial expression variations. The wavelet packet transform first decomposes image into low-, midand high-frequency subbands and the multiband feature fusion technique is incorporated to select the subbands that are invariant to illumination and expression variation separately. These subbands form the optimal feature sets. Parallel radial basis function neural networks are employed to classify these feature sets. The scores generated by the neural networks are combined by an adaptive fusion mechanism where the level of illumination variations of the testing image is estimated and the weights are assigned to the scores accordingly. The experimental results show that DOMF outperforms other algorithms and also achieves promising performance on illumination and facial expression variation conditions