We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
An Efficient Robust Eye Localization by Learning the Convolution Distribution Using Eye Template.
- Authors
Li, Xuan; Dou, Yong; Niu, Xin; Xu, Jiaqing; Xiao, Ruorong
- Abstract
Eye localization is a fundamental process in many facial analyses. In practical use, it is often challenged by illumination, head pose, facial expression, occlusion, and other factors. It remains great difficulty to achieve high accuracy with short prediction time and low training cost at the same time. This paper presents a novel eye localization approach which explores only one-layer convolution map by eye template using a BP network. Results showed that the proposed method is robust to handle many difficult situations. In experiments, accuracy of 98% and 96%, respectively, on the BioID and LFPW test sets could be achieved in 10 fps prediction rate with only 15-minute training cost. In comparison with other robust models, the proposed method could obtain similar best results with greatly reduced training time and high prediction speed.
- Subjects
FACIAL expression; SIGNAL convolution; BACK propagation; ROBUST control; PREDICTION models
- Publication
Computational Intelligence & Neuroscience, 2015, Vol 2015, p1
- ISSN
1687-5265
- Publication type
Article
- DOI
10.1155/2015/709072