Preprocessing and Feature Sets for Robust Face Recognition

Many face recognition algorithms have been developed over the past few years but the problem remains challenging, especially for images taken under uncontrolled lighting conditions. We show that the robustness of several popular linear subspace methods and of Local Binary Patterns (LBP) can be substantially improved by including a very simple image preprocessing stage based on gamma correction, Difference of Gaussian filtering and robust variance normalization. LBP proves to be the best of these features, and we improve on its performance in two ways: by introducing a 3-level generalization of LBP, Local Ternary Patterns (LTP), and by using an image similarity metric based on distance transforms of LBP image slices. We give experimental results on two face recognition sets chosen for their difficult lighting conditions: version 1 of the Face Recognition Grand Challenge experiment 4, and the full Yale-B dataset.