This paper proposes the use of Extreme Learning Machine

This paper proposes the use of Extreme Learning Machine regression
framework to generate virtual frontal view from its corresponding side view. Kernel version of ELM is used for the non-linear mapping
estimation between frontal and its corresponding non frontal view. Non Linear
estimation using proposed KELM is also found to be satisfactory even in the case
of compressed images. In addition to regression framework, this paper also
explored few popular face features like LBP, HOG, appearance and entropy
features, with ELM classification framework. Combination of appearance and
entropy feature set is found to give best classification accuracy among the
considered feature set. Proposed experiments are conducted on subset of FERET
database with images upto 45° pose variation.

Keywords: ELM(Extreme
Learning Machine),KELM(Kernel Extreme Learning Machine).

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

1  
Introduction

During last few
years, research for accurate and reliable face recognition has made
considerable progress.
Machine vision face recognition depends largely on features extracted from eye,
nose and mouth regions of the input face. These
discriminant features becomes occluded in case of face pose like profile pose.
Recognition capability degrades further due to variability in illumination and
scales. To achieve the objective of robust face recognition across poses,
various approaches like: pose normalization, pose adaptive filters and pose
synthesis has been proposed in the literature. All these approaches extract
high dimensional features (local/global) from the images and integrate them by
using learning algorithms. Extreme Learning Machine (ELM) proposed by Huang
et-al. 1,2 for single hidden layer feed- forward networks, has gained
importance among various learning algorithms. Speed advantage and convex model
structure of ELM is responsible for its usage across various domains. ELM maps
the input features to the feature space using non-linear activation function
1-3. In the mapping process of the input space RD to a high
dimensional feature space RL (ELM Space), the dimensionality L of the ELM space is usually
empirically chosen. In order to avoid the application of time-consuming
algorithms for the determination of the ELM space dimensionality, kernel
versions of the ELM classifier have been proposed (3, 6).

   The idea in kernel versions of the ELM 3,6
is to avoid the direct calculation of the network hidden layer outputs, and the
inclusion of the inherent encoding in the so-called ELM kernel matrix defined
by K = ?T?, where ? ? RL×N refers to the training data representations
in the ELM space and N is the number of training data. This
approach however contradicted the standard definition of kernel matrix.
Parviainen et al 8 used Cholesky decomposition of the kernel matrix defined
on the input training data, in order to calculate an appropriate matrix ?? RN×N. In 9 showed that ELM kernel
definition can be adopted for the calculation of the ELM kernel matrix for the
RBF and sigmoid hidden layer activation functions. For appropriate ELM space
determination they had used a low-rank decomposition of the kernel matrix
defined on the input training data. Recently, Goel et al. 9 used kernel ELM
for pose synthesis.

In this
paper we explored both regression and classification framework of ELM to
recognize faces across the poses. Non-linear regression via kernel ELM is
proposed to estimate the non-linear mapping between frontal face views from its
counter-part non-frontal views for effective face recognition. This work avoids
time consuming step for finding optimal value for hidden neurons by
experimental tuning. Compared to linear regression, KELM regression is more
efficient for virtual view generation as it considers non-linear shape of the
face view. Various features like LBP, HOG, entropy and intensity features are
explored with ELM classification framework. The rest of this paper is organized
as follows: In Section 2, we give an outline of ELM classification framework as
well as ELM regression framework for pose normalization. Section 3 presents the
outline of the proposed pose normalization approach. In Section 4 results
obtained with FERET face pose images upto 45 degree deviation are analyzed and
interpreted. Finally, in section 5 we conclude the paper and highlight some of
the future ideas for research.

x

Hi!
I'm Clifton!

Would you like to get a custom essay? How about receiving a customized one?

Check it out