Clayton School of Information Technology

Seminar series for students, Semester 1, 2008

When: Tuesday 13th May, 12noon

Where: Room 135, Building 26.

 

Topic: “The GIFT: CBIR, MRML and benchmarking”

 

Abstract:

Research in the area of content-based image retrieval (CBIR) is often hindered by the lack of common standards for CBIR systems. Frequently researchers find that they have to write code for an entire CBIR system - feature extraction, indexing and database system, query engine, user interface and communication protocols - even though they may only be interested in one aspect of CBIR from a research point of view. Not only does this slow research, it almost invariably results in tightly coupled systems with no reusable components. The GNU Image-Finding Tool  (GIFT, http://www.gnu.org/software/gift/gift.html) is an open framework for content-based image retrieval that addresses this problem, with the aim of promoting code reuse among researchers and application developers.

The GIFT explicitly allows for the addition of new query paradigms to the framework. It uses MRML (the Multimedia Retrieval Markup Language) as its communication protocol for client-server communication. MRML is XML-based and is documented at http://www.mrml.net. The use of MRML decouples the interface from the query engine, allowing researchers to use freely available interfaces if that is not their area of interest.  The converse is also true. An additional benefit of the use of MRML is that it permits the development of a common benchmarking harness for MRML-compliant CBIR systems. This is being put into practice at the Benchathlon CBIR contests (http://www.benchathlon.net/).

 

Speaker’s Biography:

Dr David Squire is a senior lecturer in the Clayton School of Information Technology. Prior to joining the Faculty of Information Technology at Monash in 2000, he was a Research Fellow in the Computer Vision Group at the University of Geneva. He has worked on invariant pattern recognition, machine learning, multimedia indexing and retrieval (including the development of the GNU GIFT CBIR system), model-based image segmentation, and plagiarism detection.