Fast label propagation on facial images using a pruned similarity matrix

Abstract

The label propagation process, which is often used to semantically annotate (tag) large amounts of multimedia data assets must be fast, in order to be efficient. In this paper, a novel facial images fast labeling method that is essentially a semi-supervised face recognition approach, is presented. The proposed method is based on the acceleration of a state of the art facial identity label propagation technique. The new method is called pruned label propagation due to the fact that the facial label inference is conducted using a similarity matrix containing fewer entries, namely the pairwise similarities that reside in the main and the off-diagonals of this matrix. Experiments conducted on facial image labeling in three stereoscopic movies, confirm the increased labeling accuracy and the reduced computational complexity of the proposed method.