Face annotation for personal photos using context-assisted face recognition

Abstract

Face annotation for personal photos has a number of potential applications in Multimedia Information Retrieval (MIR). We propose a novel face annotation method that systematically combines contextual information of photos with traditional Face Recognition (FR) technologies. To overcome the limitations of conventional FR-based annotation approaches, situation and person clustering methods are proposed based on context of personal photos. By relying on the proposed clustering methods, multiple facial images that belong to the same person are applied to subspace FR methods. This allows traditional FR methods to take the advantages of feature- or measurement-level fusion strategies for automatic annotation of faces on personal photos. The effectiveness of the proposed method was successfully tested on 1,120 personal photos selected from standard MPEG-7 Visual Core Experiment 3 (VCE-3). Extensive experimental results showed that the proposed method improves annotation accuracy by at least an order of magnitude compared with conventional FR-based approaches.