Open AccessEye centre localisation with hybrid regression framework
Author(s) -
Li Bin
Publication year - 2018
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2018.0123
Subject(s) - computer science , artificial intelligence , convolutional neural network , face (sociological concept) , computer vision , pattern recognition (psychology) , feature (linguistics) , convolution (computer science) , eye tracking , facial recognition system , feature extraction , artificial neural network , social science , linguistics , philosophy , sociology
The location of the eye is an important feature for computer vision and pattern recognition applications including psychological analysis, facial expression recognition and auxiliary driving. Most of the current work typically performs face detection first, followed by eye centre localisation. Such an algorithm fails to locate eye centre in some cases because the illuminance and external environment changes the facial features blurred in the images; therefore, the traditional face detectors cannot accurately determine the region of the face. A method for eye centre localisation which is based on convolution neural networks and regression framework is proposed. First, eye region is selected from the facial images according to convolutional neural networks. Then, the accurate eye centre is estimated by a linear regression model. Finally, the method on BioID and GI4E datasets is evaluated. The experimental result showed that the proposed algorithm has comparable or better performance compared with the state‐of‐the‐art methods.