Logarithmic Dyadic Wavelet Transform based Face Recognition

In this paper, we present a new local descriptor based on logarithmic dyadic wavelet transform. The interest points at different scales have been identified and a region of size 40x40 is considered around each interest point to obtain gradient-orientation histogram in the transformed space. The chi square distance metric is used for classification. Extensive experiments have been conducted on standard face datasets such as ORL and LFW to demonstrate the suitability of the proposed descriptor for face recognition. A comparative analysis with some of the well known methods is also provided to argue that the proposed method is comparatively perform much better than the existing methods.