Stereo facial image clustering using double spectral analysis

Abstract

In this work we proposed a new variant of spectral clustering using double spectral analysis which proved to be able to achieve better clustering results. The present work focuses on the special case of 3D videos and the implication of their use. Various improvements are introduced including the use of stereo over mono videos, the use of double spectral clustering over spectral clustering and the use of multiple representative images per trajectory for more robust trajectory representation. Extended experiments have been conducted in three 3D full feature commercial films which revealed the power of stereo face clustering in comparison with single channel face clustering.