3D retrieval system based on cognitive level: human interface for 3D building database

This paper shows a media-linkage method on cognitive level based on adjective annotation method. The method consists of mapping from still-images and 3D models to adjective texts because adjective texts as a cognitive level can be used for different multimedia data to link texts, images, 3D object models, etc. The method automatically obtains adjective scores and labels adjective texts for all the images and 3D models in the database. To construct the mapping, we have analyzed the relationship between adjectives and graphical parameters of not only street images but also 3D building models. We set up a pilot study to find 3D buildings from the database in a street image of Minato Mirai 21 (Yokohama) by using the media-linkage method.