Face Recognition via Ensemble SIFT Matching of Uncorrelated Hyperspectral Bands and Spectral PCTs
Mohd Fairul Mohd-Zaid · 2012
Face recognition through hyperspectral images is a concept which is still in its infancy. Although the conventional method of face recognition using Red-Green-Blue (RGB) or grayscale images has been advanced over the last twenty years, these methods are still shown to have weak performance whenever there are variations or changes in lighting, pose, or temporal aspect of the subjects. A hyperspectral representation of an image captures more information that is available within a scene than a RGB image therefore it is beneficial to study the performance of face recognition using a hyperspectral representation of the subjects’ faces. The results from a variety of methods of performing face recognition using the Scale Invariant Transformation Feature (SIFT) algorithm as a matching function on uncorrelated spectral bands, principal component representation of the spectral bands, and the ensemble decision of the two were studied. We conclude that there is no dominating method in the scope of our research; however, we do obtain three algorithms that outperform the results obtained from a previous study which only considered a SIFT application on a single hyperspectral band, and our algorithm that utilizes fusion performs very well under temporal variation as supported by the findings in our results. The novelty of this research is the application of SIFT on hyperspectral face images as a matching function in performing face recognition which has never been performed prior to this study.