Masters Theses

Author

Weizhi Meng

Abstract

"Nowadays people are more interested in searching the relevant images directly through search engines like Google, Yahoo or Bing, these image search engines have dedicated extensive research effort to the problem of keyword-based image retrieval. However, the most widely used keyword-based image search engine Google is reported to have a precision of only 39%. And all of these systems have limitation in creating sentence-based queries for images.

This thesis studies a practical image search scenario, where many people feel annoyed by using only keywords to find images for their ideas of speech or presentation through trial and error. This thesis proposes and realizes a sentence-based image search engine (SISE) that offers the option of querying images by sentence. Users can naturally create sentence-based queries simply by inputting one or several sentences to retrieve a list of images that match their ideas well.

The SISE relies on automatic concept detection and tagging techniques to provide support for searching visual content using sentence-based queries. The SISE gathered thousands of input sentences from TED talk, covering many areas like science, economy, politics, education and so on. The comprehensive evaluation of this system was focused on usability (perceived image usefulness) aspect. The final comprehensive precision has been reached 60.7%. The SISE is found to be able to retrieve matching images for a wide variety of topics, across different areas, and provide subjectively more useful results than keyword-based image search engines"--Abstract, page iii.

Advisor(s)

Shi, Yiyu

Committee Member(s)

Choi, Minsu
Fan, Jun, 1971-

Department(s)

Electrical and Computer Engineering

Degree Name

M.S. in Computer Engineering

Publisher

Missouri University of Science and Technology

Publication Date

Fall 2015

Pagination

viii, 30 pages

Note about bibliography

Includes bibliographic references (pages 28-29).

Rights

© 2015 Weizhi Meng, All rights reserved.

Document Type

Thesis - Open Access

File Type

text

Language

English

Library of Congress Subject Headings

Querying (Computer science) -- Design
Search engines
Digital images
Pattern recognition systems
Semantics

Thesis Number

T 10794

Electronic OCLC #

936208670

Share

 
COinS