Face recognition using SIFT descriptors extracted from multiresolution images

In this work, we developed a technique for face recognition using the idea of multiresolution face recognition. The multiresolution subbands are generated by using discrete wavelet transform (DWT). We then apply scale invariant feature transform (SIFT) to extract the salient feature descriptors at each subband using the resulting low frequency subband of the image. The descriptors are used to perform the recognition of the faces in each subband with different resolutions. Then decisions coming from each subband are combined by using simple majority voting to increase the recognition performance. Proposed, multiresolution SIFT approach shows promising results and outperforms the conventional SIFT approaches.