Abstract
In this paper we discuss how to create an image database with a good user interface using semantic vectors. The interface realizes content-based associative retrieval without using a thesaurus, and hypertext-like retrieval without linking terms explicitly. Semantic vectors, which represent one-layered semantic networks as fixed-dimensional vectors, are constructed from encyclopedia texts. The database includes about 170 images and each of them are attached by the caption which describes the image in several words. We emphasize the effectiveness of indexing and retrieving images using semantic vectors through some experiments.
