大規模顔画像データベースに対する印象語による類似顔画像検索技術に関する研究

概要

論文の詳細を見る
Multimedia information retrieval has been addressed as one of the most promisingsolutions for accessing rapidly growing multimedia data. Traditional multimediainformation retrieval systems have dealt with the same kind of media data. Withdiversification of multimedia data, however, it becomes necessary to develop multimediainformation retrieval technology among different media data. In order to achieve thepurpose above, we have developed the cross-media information retrieval system betweentext and images. Based on linguistic information provided by associated text and visualinformation provided by image features, an image retrieval system has the capable ofefficiently searching appropriate image data from huge image database. In this paper, asthe application systems for the cross-media information retrieval system between languagekeyword and image, we introduce similar faces retrieval systems by impression keywords.This system will utilize for specifying the criminal by using the witness's impression towardthe criminal. Since impressions of the witness are created by seeing the criminal only for amoment, their impressions are very ambiguous comparing the detailed impression datawhich the police stations have stored. Then, we propose the method to solve thedifference between two kind of impressions, and develop the system which is able toretrieve face images suitable to a given impression keyword.
徳島大学の論文