Image retrieval by web context: Filling the gap between image keywords and usage keywords

Abstract

The Web contains a wide variety of images published by the general public. Because the image is exposed as part of a Web page, the image is specified by not only the image content, but also the Web content using the image. This generates a gap between the content keywords and the usage keywords of the image. To fill the gap, we propose a concept of "context" of images in the Web. The context of an image is the set of Web contents (e.g. text and images) surrounding the image. We extend the notion of surroundings to include Web contents associated with the image by Web document structure and hyperlinks. We define different types of contexts for an image in accordance with types of associations between the image and the surrounding content. By visualizing context of each image, users can distinguish reliable image content among similar images based on the usage and/or reputation of the image. In this paper, we propose methods for searching and visualizing the context of images. We also propose a method for retrieving images based on their context. © Springer-Verlag Berlin Heidelberg 2003.