Weighted Link Analysis for Logo and Trademark Image Retrieval on the Web

1
Weighted Link Analysis for Logo and Trademark
Image Retrieval on the Web

• Epimenidis Voutsakis
• Euripides G.M. Petrakis
• Evangelos Milios
• Technical University of Crete
• Dalhousie University

2
Image Retrieval on the Web

• Text queries
• Answers images in Web pages
• Relevant but not always important
• From corporate web sites, organizations
• From individuals and small companies
• Link analysis assign higher ranking to answers
  from important web site
• Important doesnt mean relevant !!

3
Link Analysis

• PageRank and HITS for text retrieval
• PicASHOW for Web pages with images
• WPicASHOW handles image and text content in
  queries and Web pages
• Main idea co-cited and co-contained images are
  likely to be related

4
Image descriptions as

• Text surrounding images in Web pages
• Image filename, Alternate text, Page title,
  Caption
• Image features
• Intensity histogram, Energy spectrum, Moment
  invariants

5
Example
6
WPicASHOW

• Queries are matched against text and image
  descriptions on links
• Create the focused sub-graph F
• Authorities principal eigenvector of
  (WI)MT(WI)M
• W page to page relationships in F
• M page to image relationships in F
• Rank answers by authority value

7
Text Queries
8
Queries by text and image
9
Conclusions

• Text retrieval accurate
• Relevant but not always important answer
• PicASHOW retrieves important but not always
  relevant answers
• WPicASHOW good compromise between relevance and
  importance
• Handles image content and queries by image
  example

10
Web Implementation

• Try WPicASHOW at
• http//www.ece.tuc.gr/intellisearch
• Over than 1.5 million pages with images
• Selection of retrieval method
• Link analysis method
• And more..