Query generation from multiple media examples

Ren, R. and Jose, J.M. (2009) Query generation from multiple media examples. In: Seventh International Workshop on Content-Based Multimedia Indexing (CBMI 2009), Chania, Crete, 3-5 June 2009, pp. 138-143. (doi: 10.1109/CBMI.2009.13)

Abstract

This paper exploits an unified media document representation called feature terms for query generation from multiple media examples, e.g. images. A feature term refers to a value interval of a media feature. A media document is therefore represented by a frequency vector about feature term appearance. This approach (1) facilitates feature accumulation from multiple examples; (2) enables the exploration of text-based retrieval models for multimedia retrieval. Three statistical criteria, minimised chi-squared, minimised AC/DC rate and maximised entropy, are proposed to extract feature terms from a given media document collection. Two textual ranking functions, KL divergence and a BM25-like retrieval model, are adapted to estimate media document relevance. Experiments on the Corel photo collection and the TRECVid 2006 collection show the effectiveness of feature term based query in image and video retrieval.

• Item Type: Conference Proceedings
• Keywords: Feature term, query generation, aggregation model, media retrieval.
• Status: Published
• Refereed: Yes
• Glasgow Author(s) Enlighten ID: Jose, Professor Joemon and Ren, Dr Reede
• Authors: Ren, R., and Jose, J.M.
• Subjects: Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4050 Electronic information resources; Q Science > QA Mathematics > QA76 Computer software
• College/School: College of Science and Engineering > School of Computing Science
• Research Group: information retrieval
• Copyright Holders: Copyright © 2009 IEEE
• First Published: First published in Seventh International Workshop on Content-Based Multimedia Indexing (CBMI 2009)
• Publisher Policy: Reproduced in accordance with the copyright policy of the publisher