Personal photo organization using event annotation

Photo events in personal photo collection management are important because people take photos according to the specific events. Event is one of important cues to retrieve photos which are well remembered by people. However, event classification work needs some useful cues to annotate the photo events well. Our main contribution is to annotate photo events automatically using simple temporal cues such as month, weekday/weekends, day/night, the correlations between scene and events, and the relation between concepts such as people/non-people and indoor/outdoor and events. In this paper, we propose an automatic photo event classification framework to address these problems. The framework is composed of four main components: the temporal detector, the scene detector, the concept detector, and the event detector. Our experimental results show that the methods in the framework are useful to detect the events of personal photos. According to the contents of events, we now can detect four events: beachfun, graduation, urbantour, and yardprk, and are expanding to detect other events.