Content-Based Image and Video Retrieval (Dagstuhl Seminar 02021)
Hans‐Peter Kriegel, Jitendra Malik, Linda G. Shapiro, Remco C. Veltkamp · DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · 2002
Content-based Image Retrieval is a complex field involving various aspects.We present here the developments made since 5 years at the University of Geneva.In particular, our work has led to the release of the GIFT platform for CBIR.The emphasis is placed on a flexible architecture allowing transparent extensions in various directions.To quantify the performances of this basic system, we have investigated the way of performing objective evaluation.In particular, we are actively participating in the development of the Benchathlon framework.We also present extensions and continuation of our work in the direction of learning for semantic feature simulation, multimedia (and multimodal) processing.One important side application we define is multimedia document annotation that we think will be crucial in many tasks such as learning and evaluation. Human image perception and shape retrieval John P. Eakins, University of Northumbria at NewcastleShape retrieval still remains an intractable problem.Through projects such as ARTISAN and SPIRIT we have tried to tackle this problem by developing retrieval techniques based on models of human shape perception.Our prototype ARTISAN shape retrieval systems have achieved some measure of success through implementing rules based on Gestalt principles to group components into perceptually significant regions for matching.Analysis of retrieval failures has led us to propose new matching techniques based on multiple views of an image.Possible ways of implementing these techniques are discussed.