Loading…
A Dimensional Emotion Analysis Algorithm Based on Feature Reuse Mechanism
Dimensional emotion recognition research is an important branch of Affective Computing. However, it is challenging due to the emotional gap between emotion and audio-visual features. Motivated by the powerful feature learning ability of deep neural networks, this paper proposes a dimensional emotion...
Saved in:
Main Authors: | , |
---|---|
Format: | Conference Proceeding |
Language: | English |
Subjects: | |
Online Access: | Request full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Dimensional emotion recognition research is an important branch of Affective Computing. However, it is challenging due to the emotional gap between emotion and audio-visual features. Motivated by the powerful feature learning ability of deep neural networks, this paper proposes a dimensional emotion analysis algorithm based on feature reuse mechanism. Our goal is to predict the continuous values of the emotion dimensions from visual modality. In this work, we proposed two different Convolutional Neural Networks (CNNs) based on feature-reuse mechanism for dimensional emotion recognition. Firstly, directly-reuse CNN (DRCNN) is proposed with the single layer of feature reuse. Secondly, we proposed a model that connects each layer to every other layer in a feed-forward fashion which can be considered as the feature-maps of all preceding layers of the current layer. And its own feature-maps fusion strategy and the result of fusion is used as the input of the next layer. We call the network as fully-reuse CNN (FRCNN). We evaluate our proposed architecture on the AVEC2012 micro expression dataset. The DRCNN obtains significant improvements over the state-of-the-art, and FRCNN achieves the best performance. |
---|---|
ISSN: | 2164-5221 |
DOI: | 10.1109/ICSP.2018.8652467 |