A fundamental consideration for identification of objects in images without use of text tags

Recently, several image retrieval systems have been developed, which regard the Web as a large scale image database. In this study, we use the image itself as index keys. From the Web, a set of similar images is retrieved based on numerical characteristics of images and simultaneously text information accompanying with the images is presented to a user. The user evaluates the relationship between the images and the presented text information and it is fed back to the system for improvement. This paper presents a basic scheme of the proposed system and a prototype system for the selection of the car´s shape.