PhD Dissertation

Semi-Automatic Person-Annotation in Context-Aware Personal Photo Collections


Recent years have seen a revolution in photography with a move away from analogue film capture towards digital capture technologies, resulting in the accumulation of large numbers of personal digital photos. This means that people now have very large collections of their own personal photos, which they must manage and organise.

In this thesis we present a prototype context-aware photo management system called MediAssist, which facilitates browsing, searching and semi-automatic annotation of personal photos. We propose an approach to semi-automatic person-annotation in personal photo collections that facilitates the annotation of people in personal photo collections in a batch manner, by suggesting annotations to users as they interact with the system. We propose person classification and retrieval techniques based on analysis of the context of photo capture in addition to analysis of the image content of the photo. We use classification techniques to suggest names for faces detected in photos, and retrieval techniques suggest faces for a query name. We implement the proposed techniques and integrate them into the interface of the MediAssist prototype photo management system.

We provide a comprehensive empirical study of the proposed person classification and retrieval techniques, using the real photo collections of a number of users. We also conduct a user study which confirms the effectiveness of the semi-automatic person-annotation approach, and the utility of the system for real users when used as part of a photo management system.

Supervisor: Prof. Alan Smeaton

External Examiner: Prof. Susanne Boll