Automatic organization of large photo collections

Modern digital photography allows users to capture, store, and share thousands of digital photographs at one time. As a result, simply browsing the photo collection becomes a daunting task. A user must see and deal with every single photograph in the collection. Tasks related to browsing, such as searching for a specific photograph, or choosing a few photographs to share become difficult. Organizing the photographs to exploit this organization is one way to simplify these tasks; a user may take advantage of the organization when carrying out any of the above tasks. Unfortunately organizing the photographs by hand often requires more effort than users will apply. In this dissertation I show how using cues from metadata and image content, large collections of photographs can be organized. The photograph collection is automatically partitioned into a tree of related "events" and then a single photograph for each event can be selected to represent that group. For any given node of the tree, the user is shown only the representative photographs from the children of the node, thus reducing the visual information that they must deal with at any one time. Browsing the photographs is equivalent to traversing the tree. Other interactions with the photograph (e.g. tagging) can be carried out on individual photographs or sub-trees. The methods that I developed were informed by two user studies. The first study shows that representative photographs exist in large collections of photographs, and that humans are able to perform such selection. The second study helps illuminate the process that humans carry out when asked to select a representative photograph. The findings of these user studies helped inform the development of new methods for automatic selection of representative photographs. I present a full implementation of these methods. The implementation allows a user to browse, tag, and search photographs either on a desktop PC or over the World Wide Web.