Please use this identifier to cite or link to this item:
Title: Multilingual information retrieval and query expansion by text and image features
Authors: Zhou, Hong.
Keywords: DRNTU::Library and information science::Libraries::Information retrieval and analysis
Issue Date: 1997
Abstract: With the widespread use of multilingual and multimedia information, there is a pressing need to efficiently manage, store, manipulate and retrieve these information in a wide spectrum of applications. This thesis presents an approach in implementing intelligent information retrieval systems and studying the effects of expanding initial text queries using raw image features. We first construct a multilingual information system which combines both image and text retrieval on the World Wide Web. It has a novel user interface that can accept queries expressed in English, Chinese and mixed text. We then build up a large image data collection with relevance judgement and standard query set. Based on that, we investigate the effects of expanding initial text queries using colour, greyscale and texture features. Extensive experiments are performed in a two-pass retrieval by using the different features, and the results were compared using the recall-precision measure. Our results show that while raw image features perform poorly when used on their own, they increase the average precision more significantly than text annotations in query expansion. Moreover, the findings hold at all precision levels, and are not sensitive to the image features used and acquisition parameters of the image features. Subsequently, we provide the possible explanations by quantitative and qualitative analyses. The background theories in information retrieval such as ranking model and relevance feedback, and research issues in feature-based image retrieval such as indexing and similarity measure are also reviewed. Besides, other approaches in combination of image and text features are studied as well.
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:SAS Theses

Files in This Item:
File Description SizeFormat 
  Restricted Access
20.3 MBAdobe PDFView/Open

Page view(s) 50

Updated on Nov 24, 2020

Download(s) 50

Updated on Nov 24, 2020

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.