Equivalent Relationship of Feedforward Neural Networks and Real-Time Face Detection System

Authors:
Shuzhi Sam Ge;Yaozhang Pan;Qun Zhang;Lei Chen
Affiliations:
Social Robotics Lab, Interactive and Digital Media Institute, Department of Electrical and Computer Engineering, National University of Singapore 117576;Social Robotics Lab, Interactive and Digital Media Institute, Department of Electrical and Computer Engineering, National University of Singapore 117576;Social Robotics Lab, Interactive and Digital Media Institute, Department of Electrical and Computer Engineering, National University of Singapore 117576;Social Robotics Lab, Interactive and Digital Media Institute, Department of Electrical and Computer Engineering, National University of Singapore 117576
Venue:
Proceedings of the FIRA RoboWorld Congress 2009 on Advances in Robotics
Year:
2009

Citing 7
Cited 0

Approximation capabilities of multilayer feedforward networks

Neural Networks
Original Contribution: Approximation of continuous functions on Rd by linear combinations of shifted rotations of a sigmoid function with and without scaling

Neural Networks
Detecting Faces in Images: A Survey

IEEE Transactions on Pattern Analysis and Machine Intelligence
Robust Real-Time Face Detection

International Journal of Computer Vision
Weighted locally linear embedding for dimension reduction

Pattern Recognition
Capabilities of a four-layered feedforward neural network: four layers versus three

IEEE Transactions on Neural Networks
Kernel orthonormalization in radial basis function neural networks

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, we mainly investigate a fast algorithm, Extreme Learning Machine (ELM), on its equivalent relationship, approximation capability and real-time face detection application. Firstly, an equivalent relationship is presented for neural networks without orthonormalization (ELM) and orthonormal neural networks. Secondly, based on the equivalent relationship and the universal approximation of orthonormal neural networks, we successfully prove that neural networks with ELM have the property of universal approximation, and adjustable parameters of hidden neurons and orthonormal transformation are not necessary. Finally, based on the fast learning characteristic of ELM, we successfully combine ELM with AdaBoost algorithm of Viola-Jones in face detection applications such that the whole system not only retains a real-time learning speed, but also possesses high face detection accuracy.