Applications of transfer learning for convolutional neural networks (CNNs) have shown to be an efficient alternative for solving recognition tasks rather than designing and training a new neural network from scratch. However, there exists several popular CNN architectures available for various recognition tasks. Therefore, choosing an appropriate network for a specific recognition task, specifically designed for a humanoid robotic platform, is often challenging. This study evaluates the performance of two well-known CNN architectures; AlexNet, and VGG-Face for a face recognition task. This is accomplished by applying the transfer learning concept to the networks pre-trained for different recognition tasks. The proposed face recognition framework is then implemented on a humanoid robot known as NAO to demonstrate the practicality and flexibility of the algorithm. The results suggest that the proposed pipeline shows excellent performance in recognizing a new person from a single example image under varying distance and resolution conditions usually applicable to a mobile humanoid robotic platform.
[1]
Masakazu Matsugu,et al.
Subject independent facial expression recognition with robust face detection using a convolutional neural network
,
2003,
Neural Networks.
[2]
Xiaogang Wang,et al.
DeepID3: Face Recognition with Very Deep Neural Networks
,
2015,
ArXiv.
[3]
Tao Wang,et al.
End-to-end text recognition with convolutional neural networks
,
2012,
Proceedings of the 21st International Conference on Pattern Recognition (ICPR2012).
[4]
Geoffrey E. Hinton,et al.
ImageNet classification with deep convolutional neural networks
,
2012,
Commun. ACM.
[5]
Jason Weston,et al.
A unified architecture for natural language processing: deep neural networks with multitask learning
,
2008,
ICML '08.
[6]
Andrew Zisserman,et al.
Deep Face Recognition
,
2015,
BMVC.