Related Experiment Videos
Subject independent facial expression recognition with robust face detection using a convolutional neural network.
Masakazu Matsugu1, Katsuhiko Mori, Yusuke Mitari
1Canon Research Center, 5-1, Morinosato-Wakamiya, Atsugi 243-0193, Japan. matsugu.masakazu@canon.co.jp
Summary
This study presents a novel facial expression recognition model capable of reliably detecting smiles across diverse individuals and appearances. The algorithm achieves high accuracy, demonstrating subject independence and robustness for advanced human-computer interaction.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Human-Computer Interaction
Background:
- Facial expression recognition is crucial for perceptual user interfaces.
- Existing methods struggle with individual variability and appearance changes.
- Subject independence and invariance to transformations are key challenges.
Purpose of the Study:
- To develop a robust facial expression recognition algorithm.
- To achieve subject independence and invariance to translation, rotation, and scale.
- To enable reliable smile detection in diverse conditions.
Main Methods:
- A rule-based algorithm for facial expression recognition.
- Robust face detection using a convolutional neural network.
- Integration of visual cues and saliency scores for discrimination.
Main Results:
- Achieved a 97.6% recognition rate for smiles on 5600 images from over 10 subjects.
- Demonstrated subject independence and robustness to appearance variations.
- Successfully discriminated smiling from talking using visual cue voting.
Conclusions:
- The proposed model offers reliable facial expression recognition with subject independence.
- It is the first model to combine subject independence with robustness to appearance variability.
- This work advances the development of autonomous perceptual user interfaces.