Name : 2004 ITE Annual Convention
Location : [in Japanese]
Date : August 25, 2004 - August 27, 2004
In this paper, we propose a novel categorization of web images into various categories based on their text and image data, and propose new applications of the categorization. We define nine categories of images used in WWW and make a tree structure for the categorization and apply SVM to each stage of the categorization. We conduct experiments to categorize a large amount of web images collected from chosen 130 web sites of listed companies of Japan and they are well categorized with 90 % precision. We also show new applications by using the image categories to help users to navigate web pages.
