A Novel Feature Set for Video Emotion Recognition
In video recommendation systems, emotions are used along with several other proposed content-based video features. However, such features are independently based on visual or audio signals and the relationship representing the dependencies between the visual and the audio signals is still unexplored. In order to solve this problem, a novel feature set called HHTC features based on the combination of Hilbert—Huang Transform (HHT) based visual features, HHT-based audio features, and cross-correlation features is proposed in this paper. In addition to the dependencies between the visual and the audio signals, the proposed HHTC features have the ability to indicate the time-varying characteristics of these signals. The proposed features are applied to video emotion recognition with the Support Vector Regression (SVR) with potential use in video affective recommendation systems. Experimental results demonstrate that the proposed approach can achieve an improved performance of video affective recognition.
S. Mo et al., "A Novel Feature Set for Video Emotion Recognition," Neurocomputing, vol. 291, pp. 11 - 20, Elsevier, May 2018.
The definitive version is available at https://doi.org/10.1016/j.neucom.2018.02.052
Intelligent Systems Center
Second Research Center/Lab
Center for High Performance Computing Research
Keywords and Phrases
Feature extraction; Recommender systems; Speech recognition; Affective recognition; Content analysis; Cross correlations; Emotion recognition; Empirical Mode Decomposition; Huang transform; Support vector regression (SVR); Time-varying characteristics; Mathematical transformations; Article; Content analysis; Correlation analysis; Emotion; Priority journal; Problem solving; Recognition; Support vector machine; Videorecording; Cross-correlation; Hilbert-Huang transform; Video affective content analysis
International Standard Serial Number (ISSN)
Article - Journal
© 2018 Elsevier, All rights reserved.
01 May 2018