Improvement of Mutual Retrieval of Visual and Audio Materials Based on Impression

Abstract

This paper revises the system mutually retrieving visual and audio materials based on impression. Visual (audio, respectively) materials are retrieved only by specifying an audio (visual) material. Visual (audio) materials having similar impression to an audio (visual) material are retrieved. To this end, the scores of the factors representing the impression of visual and audio materials are estimated, and used. Pictures are used as visual materials. This paper revises the feature values used in the estimation, and the estimation methods. The revised method is experimentally evaluated. The evaluation result shows that the precision of retrieval is optimistically good.