Abstract
We developed a system which enables users to register a large number of 2D images efficiently and to retrieve the image which matches his or her wish. In this system, the user can register, modify and delete the image data together with attributes through a Web browser, since CGI programs which the Web server executes access the image database. An experimental result shows that this method reduced the registration time to about l/7 against the traditional way of user's inputting SQL sentences. For image retrieval, we attempted to construct the category structure according to which the image data are classified. Furthermore we developed experimentally a retrieval system for images of tray (or "Bon") and nests of boxes (or "Jubako"), and verified that the user can retrieve the expected images easily. We also use the function of similar image retrieval to detect the duplicate image registration, and to help the more detailed image retrieval.
