Photo Indexing Using Person-Based Multi-feature Fusion with Temporal Context

Abstract

A photo indexing is presented to facilitate the browsing of a large collection of photos associated with specified persons in photo. The proposed method aims at exploiting additional person-identity features in order to cope with the limitations of conventional person-based image or photo indexing using facial features only. Person-identity feature is created by the combination of face region and associated contextual information. For consistent use of associated contextual information, situation-based photo clustering is adopted. Experiment was performed with 1120 generic home photos (the official database of MPEG-7 VCE-3.