The problem of the image retrieval system using kansei information has been a development of the method of connecting with objective kansei information to the image data, when the image data was appended. To solve this problem, the technique by which kansei information is mechanically extracted from the apparel monochrome pattern and the extraction system are proposed. In this technique, the shape and the position of object of pattern are recognized by the image processing. Then, the composition elements by which characteristic of pattern is shown are guessed automatically. The kansei information hence can be calculated with those values by kansei information models and are output.
The evaluation experiment was done by using ARSKI (Apparel pattern Retrieval System using Kansei Information), to proof of the effectiveness of this technique. As a result, the retrieval hit rate to 280 kinds of patterns was 66% and the effectiveness was confirmed.