Retrieval of Relevant Parts of Document Images Based on 2D Density Distributions of Characters

This paper presents a new method of document image retrieval that is capable of spotting parts of document images relevant to users’ queries. This enables us to improve effectiveness and usability of retrieval, since users are relieved from burdens of finding relevant parts in retrieved documents. The proposed method is based on the assumption that parts of document images which densely contain characters in queries are relevant to them. For the purpose of ranking relevant parts, two-dimensional density distributions of characters are calculated based on layout features such as locations of characters and distance to the nearest characters. Based on the experimental results of retrieving Japanese newspaper articles, it is shown that the proposed method is superior to a method without a function of retrieving the parts.