Local wavelet analysis for face recognition

In this paper we investigate the benefits of using local wavelet analysis to the face recognition problem. We examine two possible approaches to perform local wavelet analysis. In the first approach, discrete wavelet transform is performed on the entire face image and then the transformed image is partitioned into non-overlapping rectangular blocks. In the second approach, as in the JPEG2000 standard, the input face image is first partitioned into non-overlapping rectangular blocks, and then on each block discrete wavelet transformation is performed. Proposed approaches are tested against the occlusion problem using the AR face database and significant improvements are observed in the face recognition performance.