文献类型：期刊文献

英文题名：Emotion-wise feature interaction analysis-based visual emotion distribution learning

作者：Zhang, Jing[1];Qin, Qiuge[1];Liu, Xinyu[1];Ye, Qi[1];Du, Wen[2]

机构：[1]East China Univ Sci & Technol, Dept Comp Sci & Engn, Shanghai, Peoples R China;[2]DS Informat Technol Co Ltd, Shanghai, Peoples R China

年份：2024

卷号：40

期号：3

起止页码：1359

外文期刊名：VISUAL COMPUTER

收录：;EI(收录号：20231714026673);WOS:【SCI-EXPANDED(收录号:WOS:000979483600004)】；

基金：This study was funded by the Nature Science Foundation of Shanghai (Grant Number 22ZR1418400).

语种：英文

外文关键词：Visual emotion distribution learning; Feature interaction analysis; Interclass relationships

摘要：Image emotion is not exclusive, which makes emotion distribution learning more meaningful than emotion classification for visual emotion recognition. Consider the emotion correlations implicit in complex images do not strictly follow the universal psychological laws, which is essential for image sentiment analysis. We propose a novel emotion-wise feature interaction analysis (EFIA) method to study the emotion correlations for emotion distribution learning. It facilitates the interaction of specific features categories to learn complicated and specific inter-class relationships from the emotion feature perspective. In addition, we propose a distribution-oriented multi-task learning method to obtain a specialized distribution learning model. Experiments on public emotion datasets illustrate that our proposed method can achieve excellent performance on image emotion distribution learning and outperform most state-of-the-art methods.