Face Recognition Template in Photo Indexing: A
Proposal of Hybrid Principal Component Analysis
and Triangular Approach (PCAaTA)
L. G. Vu
1
, Abeer Alsadoon
1
, P.W.C. Prasad
1
, A. Monem
2
, A. Elchouemi
3
1
School of Computing and Mathematics, Charles Sturt University, Sydney, Australia
2
Computer Science Department, University of Technology, Baghdad, Iraq
3
Hewlett Packard Enterprise
Abstract - Many features of human faces play important roles in
designing a method that can perform face recognition. There are
many current studies in the area of face recognition, but most of
them have limitations. The Principal Component Analysis (PCA) is
one of the best facial recognition algorithms. However, there are
some noises that could affect the accuracy of this algorithm. The
PCA works well with the supports of preprocessing steps such as
illumination reduction, background removal and color conversion.
Some current solutions have showed result in using combination of
PCA and preprocessing steps. This paper proposes a hybrid
solution in face recognition using PCA as the main algorithm with
the support of triangular approach in face normalization in order
to enhance the accuracy. To evaluate, the proposed approach is
tested and the results are compared with the current solutions.
Keywords: face recognition, face detection, photo indexing,
Principal Component Analysis, triangular approach
I. . INTRODUCTION
With the rapid development of Information and
Communication Technology (ICT), there has been a
significant increase in the need of secure and convenient tools
that can be used in real life. It is difficult to retrieve the large
numbers of images and videos from mobile devices and
private computers, therefore there is a strong need to use
indexing methods.
One of the possible solutions is face recognition that was
developed by Woodrow W. Bledsoe in the 1960s. The idea is
to determine features, such as mouths, noses, eyes and ears on
images before calculating ratios and distances to a common
point that is used to compare with reference data [1].
The problem with the current available solutions is the
measurement and location progress, as they have to be done
manually. Some studies have carried out those tasks
automatically by applying many algorithms namely, Principal
Components Analysis (PCA), Linear Discriminant Analysis
(LDA) and Elastic Bunch Graph Matching (EBGM) [2].
However, those methods still have many limitations and need
to be supported by a variety of algorithms and techniques in
order to enhance the accuracy [3]. For example, one of the
latest algorithms can detect multiple faces in a color image
shows the accuracy of just 93% [4].
This paper aims to use an enhanced face recognition technique
to identify an individual and index photos and videos.
The paper is organized as follows: Section I is introduction,
section II discusses the related work on the project topic,
followed with impact factors in section III. The proposed
method and are given in section IV. The results and discussion
given in section VI followed by conclusion in the section 6
with possible future work.
II. RELATED WORK
A. Preprocessing
Preprocessing is the first steps in face recognition which
reduce noise affecting the human face images. Two main
types of noises are illumination, and background.
1. Illumination reduction
The negative effect of illumination on face detection is very
clear. Illumination leads to shape of shadows, changes in
highlights, shifts in the location and reversal of contrast
gradients [5]. Therefore, reducing illumination effects is very
important to enhance the accuracy rate. The color information
is not used in identifying edges or other features [6]. Thus,
color can be considered as noise and RGB (Red-Green-Blue)
images should be converted into YCbCr (greyscale) images.
Another illumination reduction method that can be considered
as illumination-insensitive image integral normalized gradient
image (INGI) [7].
2. Human face cropping in largescale image
The face detection and recognition can be affected in term of
accuracy and time-consuming due to large scale of
background. For instance, an image contains 90% of human
face and 10% of background is more easy and informative to
be analyzed, compared to an image contains 10% of human
face and 90% of background. The color-based segmentation is
based on the color of input image to differentiate the skin
color region against non-skin color regions [4].
B. Face detection
1. Current face detection methods
Face detection is aimed at detect and extract face feature from
a given image. One of the most popular algorithms that has
been used for a long time is the Viola and Jones algorithm [8].
This method is used to classify whether an image contains a
human face and if so, where it is. The Viola-Jones algorithm
177 978-1-4673-9919-7/16/$31.00 ©2016 IEEE SSIAI 2016