IMAGE DATABASE MANAGEMENT USING SIMILARITY PYRAMIDS A Thesis

Chen, Jau-Yuen, Ph.D., Purdue University, May, 1999. Image Database Management using Similarity Pyramids. Major Professor: Charles A. Bouman. In this work, four major components of image database have been examined: image similarity, search-by-query, browsing environments, and user feedback. We first present a formal framework for designing image similarity and search algorithms. This framework is based on a multiscale representation of both the image data and the associated parameter space. We also define a general form for the distance function which insures that branch and bound search can be used to find the globally optimal match. We then exploit the techniques of tree structured vector quantization (VQ), branch and bound search, and the triangle inequality to speed-up the search of large image databases. Our method can reduce search computation required to locate images which best match a query image provided by a user. In addition, a free parameter allows computation to be further reduced at the expense of search accuracy. Our browsing environment is based on a similarity pyramid data structure which hierarchically organizes the database so that it can be efficiently browsed. The similarity pyramid groups similar images together while allowing users to view the database at varying levels of detail. At coarse levels, the similarity pyramid allows users to view the database as large clusters of similar images. Alternatively, users can “zoom into” finer levels to view individual images. Finally, in order to include users in the “browsing loop”, we integrate relevance feedback into the browsing environment. The new browsing environment, which we call active browsing, allows users to dynamically modify the database’s organization through the definition of a set of relevant images. The relevance set is then used to adaptively prune and reorganize the pyramid to best suit the user’s task.