HIGH PERFORMANCE POSE INVARIANT FACE
RECOGNITION
*Hasan Demirel and Gholamreza Anbarjafari
Department of Electrical and Electronic Engineering, Eastern Mediterranean University
Gazimağusa, KKTC, Mersin 10, Turkey
Keywords: Face recognition, Pose invariant, histogram matching.
Abstract: A novel pose invariant face recognition system based on grey
level histogram matching is proposed. The
proposed system in this paper uses grey level histograms as feature vectors for recognition of the different
poses of faces. The process is performed by taking the cross correlation between the histogram of a test face
and the histograms of the training faces in the database. The proposed system gives 98.80% recognition rate
on the HP database of 15 face subjects. This rate is down to 92% in the case of conventional eigenfaces
method.
1 INTRODUCTION
The advances in research and development of
multimedia systems have increased the use of
biometric authentication applications such as face
recognition. The earliest work in computer
recognition of faces is reported by Bledsoe (1964),
where manually located feature points are used to
recognize a faces. Kanade (1977) used a method of
characterizing the face by geometrical
parameterisation, whereby distances and angles
between points such as eye corners, mouth
extremities, nostrils, and chin top are used.
Statistical face recognition systems, such as,
principal component analysis (PCA) based
eigenfaces method introduced by Turk and Pentland
(1977) attracted a lot of attention by the researchers.
In this method the eigenvectors of the covariance
matrix of the training set of faces are used as the
basis vectors and any test image is mapped to the
representation vector in a low dimensional
eigenspace before classification. Relhumeur et. al
(1997) introduced fisherfaces method which is based
on the linear discriminant analysis (LDA) that
minimizes the discrimination within a class and
maximizes the dircrimination between classes. Their
class specific approach provided improved
recognition results and showed robustness to
illumination changes.
Another statistical descriptor can be considered
t
o be the histogram of a gray level image which
shows the distribution in terms of occurrence
frequencies of gray level pixel intensities. Histogram
of a face image can be considered as the signature of
the face, which can be used to represent the face
image in a low dimensional space. Images with
small changes in translation, rotation and
illumination still possess high correlation in their
corresponding grey-scale histograms. This histogram
characteristic prompts the idea of using grey-scale
histograms for face detection and recognition.
There is very limited work in histogram based
m
ethods in face recognition. Yoo et. Al (1999) used
chrom
atic histograms as a model of faces. They used
histograms for the detection of faces by
backprojecting a face histogram onto the entire
image containing a face to search and detect the face
in the backprojected image. Ojala et. al. divided a
face into several blocks and then the Local Binary
Pattern (LBP) feature histograms (2002) are
extracted from each block and concatenated into a
single global feature histogram which efficiently
represents the face image. Some other work have
cooperated with histogram to recognize faces (Chen,
2004, and David et. al, 2000). The recognition of the
face is performed by simple distance based
histogram matching.
In this paper, the gray level histogram of the entire
face im
age is used as the face descriptor of a given
face. The proposed face recognition system simply
uses gray level histograms to describe faces and
recognition is achieved by incorporating histogram
282
Demirel H. and Anbarjafari G. (2008).
HIGH PERFORMANCE POSE INVARIANT FACE RECOGNITION.
In Proceedings of the Third International Conference on Computer Vision Theory and Applications, pages 282-285
DOI: 10.5220/0001078402820285
Copyright
c
SciTePress
matching by using the cross correlation between the
histogram of the input face and the histograms of the
faces in the training set. The results obtained from
the proposed histogram based system provide a
recognition rate as high as 99.60% for the ORL face
database when 5 poses of each subject out of 40
persons are used to train and remaining 5 poses are
used for performance testing. The proposed method
clearly outperforms the classical face recognition
systems such as PCA based eigenfaces and LDA
based fisherfaces methods, where this rate is down
to 78% and 87% respectively.
Additionally, due to the high correlation between
the histograms of the faces at different scales, the
proposed system is robust for scale changes. The
recognition rate for smaller faces with sizes of only
30% of the images in the training set is 98.22%,
which is within 2% reduction in the recognition rate.
This result is expected, because the information loss
measured by entropy encountered by image
shrinking down to 30% of the original image is in
the range of 2%.
2 HISTOGRAM BASED FACE
RECOGNITION
One of the methods of describing an image in lower
dimension is using histogram. Histogram of an
image can be considered as feature vector
representing of the image. In general, histogram of
an image is a statistical description of the
distribution in terms of occurrence frequencies of
pixel intensities. The size of the image histogram
depends on the number of quantization levels of the
pixel intensities. Typical monochrome image with 8-
bit representation has 256 gray levels. In a
mathematical sense, an image histogram is simply a
mapping
η
i that counts the number of pixel
intensity levels that fall into various disjoint
intervals, known as bins. The bin size determines the
size of the histogram vector. In this paper the bin
size is assumed to be 256 and the size of the
histogram vector is 256. Histogram of a
monochrome image,
η
i, meets the following
conditions
=
=
255
0i
i
N
η
(1)
where N is the number of pixels in an image. Then,
histogram feature vector, H, is defined by,
],,,[
25510
η
η
η
"=H
(2)
The similarity between two images can be
measured by using the cross correlation between the
histograms of the respective images. The maximum
correlation coefficient in the correlation vector is
taken as the measure of similarity and used in the
histogram matching process.
If H1,H2,….,HM be a set of raining face images
with different poses and M be the number of image
samples, then a given query face image, the
histogram of the query image Hq can be used to
calculate the correlation between Hq and histograms
of the images in the training samples as follows:
max(),,
iiq
HH i M
χ
==1D",
(3)
Thus, the similarity of the ith images in the
training set and the query face can be reflected
by
i
χ
, the maximum cross correlation coefficient.
The, image with the highest similarity measure, is
declared to be the identified image in the set.
The proposed system using histogram as the face
feature vector and maximum cross correlation
coefficient as the histogram matching measure is
tested on Head Pose face database (Gourier, 2004),
which contains 15 subjects with 10 selected different
poses. The face dataset is divided into training set of
n (n5) images per subject and the rest images for
the test set. The images used in the test set are not
included in the training set. The correct recognition
rates in percent are included in Table 1. Each result
is the average of 500 runs, where we have randomly
shuffled the faces in each class. The results of the
proposed system are outstanding, because even a
single image in the training set provides a correct
recognition rate as high as 94.89%. This rate is down
to 68.89% in the PCA based face recognition
systems respectively.
The proposed method shows slight improvement
as the number of training set images is increased. On
the other hand PCA based systems reaches 92%. The
results are very encouraging and the proposed face
recognition system shows a clear superiority over
the conventional face recognition systems.
Table 1: Performance of the proposed histogram based
system compared with PCA based systems.
# of Training
Images
PCA
Proposed Histogram
Matching Method
1 68.89 94.89
2 81.67 94.62
3 89.52 98.14
4 92.22 97.33
5 92.00 98.80
POSE INVARIANT FACE RECOGNITION USING IMAGE HISTOGRAMS
283
0 100 200
0
2000
4000
0 100 200
0
2000
4000
0 100 200
0
2000
4000
0 100 200
0
2000
4000
6000
0 100 200
0
2000
4000
0 100 200
0
2000
4000
6000
0 100 200
0
2000
4000
0 100 200
0
2000
4000
6000
0 100 200
0
2000
4000
6000
0 100 200
0
2000
4000
Figure 1: Faces with their PDF.
3 ROBUSTNESS ON POSE
VARIANCE
0 0.5 1 1.5 2 2.5 3 3. 5 4 4.5 5
0
10
20
30
40
50
60
70
80
90
100
Number of t raining set
Recognition Rate (%)
Hist ogram Matching
PCA
As fig. 1 shows pose differences (planer rotations)
on the images do not change the general distribution
of the grey level PDFs. The shape of the PDF is
more or less preserved, while only the amplitudes
are changing as the angle of rotation of the image is
changing.
So the shape of the histogram of a face image is
its signature in representing the image. Hence, if the
general distribution is preserved, then the correlation
of the histogram of the image with high and low
resolutions will be high.
Fig.2 shows the performance of the histogram
based face recognition system for the changing
poses. The dashed line of the figure gives the
recognition rate of the PCA system for the images
with different poses. The performance of the
proposed face recognition system is always higher
than the performance of the PCA. Obviously there is
a limit of the angle of planer rotation, where the face
image has been completely changed, e.g. only back
of head is in the image.
Figure 2: Recognition Rate (%) of images for changing
poses.
4 CONCLUSIONS
In this paper we introduced a novel face recognition
system based on grey level histogram matching.
Maximum cross correlation coefficient between the
histogram of a given face and the histograms of the
faces in the database was used for histogram
matching. 98.80% recognition rate on the HP
database of 15 face subjects was obtained by using
the proposed method while this rate was down to
92% in the case of conventional eigenfaces. It has
been shown that due to the high correlation between
the histograms of the faces at different poses, the
proposed system is robust for pose changes.
VISAPP 2008 - International Conference on Computer Vision Theory and Applications
284
REFERENCES
A. David and S. Panchanathan, Wavelet-histogram
method for face recognition, Journal of Electronic
Imaging, April 2000, Volume 9, Issue 2, pp. 217-225.
M. Turk and A. Pentland, “Face Recognition Using
Eigenfaces,” in Proc. Int. Conf. On Patt. Recog., 1991,
pp. 586-591.
N. Gourier, D. Hall, J. L. Crowley, “Estimating Face
Orientation from Robust Detection of Salient Facial
Features”, Proceedings of Pointing 2004, ICPR,
International Workshop on Visual Observation of
Deictic Gestures, Cambridge, UK.
Q. Chen, K. Kotani, F. Lee, and T. Ohmi, Face
Recognition using Region-Division VQ Histogram
Method. Signal and Image Processing (SIP04).
Relhumeur, P., Hespanha, J., and Kreigman,
D.,“Eigenfaces vs. Fisherfaces: Recognition using
Class Specific Linear Projection,” IEEE Transactions
on Pattern Anal.vsis and Machine Intelligence, 19(7),
7 11-720, 1997.
Tae-Woong Yoo, Il-Seok Oh, “A fast algorithm for
tracking human faces based on chromatic
histograms.”, Pattern Recognition Letters, vol. 20,
967-978, 1999.
T. Kanade, “Computer Recognition of Human Faces,”
Basel and Stuttgart: Birkhauser, 1977.
T. Ojala, M. Pietikainen, and T. Maenpaa,
“Multiresolution gray-scale and rotation invariant
texture classification with local binary patterns,” IEEE
Transactions on Pattern Analysis and Machine
Intelligence, vol. 24, pp. 971–987, 2002.
W.W. Bledsoe, “The Model Method in Facial
Recognition”, Panoramic Research Inc., Palo Alto,
Cal., and Rep. PRI: 15, CA, 1964.
POSE INVARIANT FACE RECOGNITION USING IMAGE HISTOGRAMS
285