객체의 속성 정보를 이용한 감성기반 이미지 검색 및 주석 기법

Sensitive Image Search and Annotation using Attribute Information of Concepts

초록

How to capture subjective aspects such as emotion, feeling from the image is still a challenging issue. In this paper, we propose a novel multi-level learning approach to predict the emotional aspects of the concept in images. The proposed approach is based on the assumption that high-level attributes of a concept will affect user’s subjectivity to the concept. For example, an animal with high-level attributes ‘small’, ‘furry’, and ‘white’ can be associated with the emotional term ‘cute’. Based on this assumption, we build a multi-layer subjectivity learning framework which consists of 1) attribute learning layer that predicts high-level attributes of a concept and 2) subjectivity learning layer which exploits the outputs from the attribute learning layer for predicting the emotional aspects of the concept. We show the strengths and potentials of our approach through various experimental results on Animals with Attributes dataset and discuss further research directions.

키워드

감성기반 이미지 검색; 감성기반 이미지 주석; 감정 분류; Sensitive image search; sensitive image annotation; emotional classification
제목
객체의 속성 정보를 이용한 감성기반 이미지 검색 및 주석 기법
제목 (타언어)
Sensitive Image Search and Annotation using Attribute Information of Concepts
저자
정진우; 이동호
발행일
2013-04
저널명
데이타베이스연구
권
29
호
1
페이지
17 ~ 36