We propose a novel categorization for web images based on their image and text data, and propose new applications using this categorization. We have defined eight categories of images used in web pages, constructed a tree structure for the categorization, and applied Support Vector Machines (SVMs) to each stage of the categorization. We then conducted experiments to categorize a large number of web images, collected from the web sites of 130 listed Japanese companies, and found that they were well categorized with 90% precision. We discuss new applications using the image categories to help users navigate web pages. In addition to this, we examined a correlation with the image categories and the classification of web pages, and conducted an experiment to construct an object from web image.
展开▼