In this paper we describe a system for storing and retrieving digital images from personal collections. The images can either be manually annotated with a set of keywords chosen by the owner of the collection, or keywords can be automatically inferred from the time and location stamps associated with the image and the Geographic Names Data Base gazetteer. User queries are matched against the image annotations using the Cosine Similarity Measure, and the best matching images are displayed. To enable browsing of the image collection, the images are clustered according to time and location, the two main factors of episodic memory. User studies were conducted to compare the two annotation systems with time and location clustering alone, using searching time, recall and precision, and user satisfaction as criteria for success. Our results show that matching based on automatically assigned keywords was significantly better than time and location-based browsing alone, and performed at least as well as matching against user-assigned annotations.
展开▼
