PDF version of this entire document
Verification tasks that are limited to geometric data (in 3-D) remain challenging. After more than a year exploring and studying this problem, it seems safe to say that diffusion-based methods are less suitable than geodesic distances for the task of identity validation and while PCA-based methods show promise, they are outperformed by GMDS and surrogate methods, at least within our experimental framework which primarily examines two publicly available datasets. Further work could explore the effect of measuring exact geodesics, combining several different classifiers, and maybe fusing photometric data.
Acknowledgements: the project was funded by the http://erc.europa.eu/European Research Council.
