Face features extraction based on multi-scale LBP

How to extract the strong features of face image is vital important in the face recognition technology. The extracted features should be robust for variation of illumination and expression. A novel feature extraction algorithm based on wavelet decomposition and LBP is proposed, which makes use of the idea of wavelet multiresolution and local characteristic of LBP. And the features extracted by this way contain holistic and local information that can be robust to identify faces. Experiment results show that the proposed method can effectively be used in face recognition with single training sample per person. The performance is better than PCA and original LBP. And the importance of different level´s lower coefficients is also analyzed.