AutoAlbum: clustering digital photographs using probabilistic model merging

Abstract

Consumers need help finding digital photographs in their personal collections. AutoAlbum helps users find their photos by automatically clustering photos into an album. The albums are presented in an easy-to-use browsing user interface. AutoAlbum uses the time and order of photo creation to assist in clustering: albums consist of temporally contiguous photos. The content based clustering algorithm is best-first probabilistic model merging, which is fast and yields clusters that are often semantically meaningful.