表題番号:2020C-663 日付:2021/04/01
研究課題Micro-macro-motion feature based multimodal method for emotion recognition
研究者所属(当時) 資格 氏名
(代表者) 理工学術院 情報生産システム研究センター 助手 田 艶玲
(連携研究者) Graduate School of Information, Production and Systems, Waseda University Doctoral Student Huilin Zhu
(連携研究者) Graduate School of Information, Production and Systems, Waseda University Doctoral Student Yanni REN
研究成果概要

Human emotion recognition has attracted a significant amount of attention from the academic community, owing to its applications in many areas including human-computer interaction, psychological studies, surveillance. However, with development with deep learning in emotion recognition, dataset plays an important role in the emotion recognition research. There are two major data problems, missing data and unlabeled data, that may affect the performance of action recognition algorithms. In addition, there is another problem that how to extract rich feature (micro-macro-motion feature) for this research. To solve these problems, three approaches are studied: 

(1) Aiming at missing data problem, a hybrid model with missing data using Quasi-Linear kernel is proposed.

(2) Aiming at unlabeled data problem, a Laplacian support vector machine (LapSVM) based semi-supervised learning is proposed, which considers the problem of learning from both labeled and unlabeled data.

       (3) Aiming at extracting rich feature difficultly, a Rich Feature Generation (RiFeGAN) method is proposed, which can generate rich feature from other modalities.