Construction of Image Retrieval System by Kansei Information Queries

In this paper, we try to construct an image retrieval system so that Kansei (a kind of the sence) information is adopted as a query of the system and pictures fitting Kansei are retrieved from image database. First, we select Kansei words from an adjective dictionary based on Factor Analysis. Next, we give relations between Kansei words and physical design elements of the pictures by Quantification Method I and fuzzy logic theory. Finally, we compare the pictures retrieved by our system with those selected by human in evaluation experiments and show the validity of our system.