Correlating Scenes as Series of Document Sentences with Images

With the popularization of the Internet and the development of multimedia technology, various data are produced and disseminated in different media such as documents, images, photographs, pictures, and movies. Therefore, there are a variety of data in different media which represent the same contents. However, the method of correlating data in different media having the same contents has not yet been established. This paper proposes a method of correlating a document with an image, both of which express the similar contents. This method bases upon the observation that a series of sentences in a document could constitute a scene, and by utilizing the similarity measure of the vector space model as a criterion, we could correlate a scene with an image if their similarity is high. An experimental result of the above-mentioned method is shown taking "The Tale of Genji" and the photographs taken on the subject of this tale, as an example. It is shown that the method works fairly good as it is intended, however there still remain difficult problems for improvement.