An Interactive Annotation Framework for Image Retrieval (夏のデータベースワークショップDBWS2004)

概要

論文の詳細を見る
Image retrieval is usually based on content information, external annotation, or a combination of the two. While content information is intrinsic to the image and thus is an objective data, external annotation may imply human and be deeply dependent on the culture and sensibility of the annotator. Consequently, two operators would not produce the same annotation for the same image, and both results may not be satisfying for user. However, an image usually contain important information impossible to retrieve from content, thus relying only on content information does not produce satisfying results. To answer to this issue, we propose an annotation framework featuring an explicit separation of objective and subjective informations, objective informations including content based information and a limited number of annotations, the other annotations being considered as subjective information and collected with the help of the user. We also propose an efficient way to apply these subjective annotations to a navigation structure calculated before-hand from the objective informations. The resulting browsing technique brings the relevance of human annotation without the cost of a before-hand annotation, and without denying the robustness of a before-hand calculated structure.
一般社団法人情報処理学会の論文
2004-07-14