Transcript Evaluation of Image Pre-processing Techniques for

Evaluation of Image Pre-processing
Techniques for Eigenface Based
Face Recognition
Thomas Heseltine
www.cs.york.ac.uk/~tomh
[email protected]
Introduction
•Growing interest in biometric authentication
•National ID cards, Airport security, Surveillance, Site access.
•Face recognition offers several advantages over other biometrics:
•Can be used without subjects knowledge.
•Human readable media.
•No association with crime, as with fingerprints.
•Data required is easily obtained and readily available.
•Approaches include:
•Neural networks, Feature analysis, Graph matching, Information
theory.
2
Principal Component Analysis
Used to create an image subspace (face space) with reduced dimensionality,
while maintaining a high level of discrimination between faces.
• A 256 by 256 pixel image of a face, represented as a vector occupies a
single point in 65,536-dimensional image space.
• Images of faces occupy a relatively small region of this large image space.
• A training set of face images is used to create a covariance matrix, from
which the eigenvectors (and eigenvalues) are computed using standard
methods.
• The eigenvectors define an image subspace, known as face space, which
maximises the spread of face images within this space.
• Selecting the eigenvectors with the highest eigenvalues (the principle
components) reduces the dimensionality of face space.
• Two face images are compared by measuring the (Euclidean) distance
between their positions in face space.
3
Eigenfaces
•The eigenvectors are sorted in order of descending eigenvalues
and the 30 greatest eigenvectors are chosen to represent face space.
•This reduces the dimensionality of the image space to 30, yet
maintains a high level of variance between face images throughout
the image subspace.
•Any face image can then be represented as a vector of coefficients,
corresponding to the ‘contribution’ of each eigenface.
Each eigenvector can be displayed as an image and due to the likeness to
faces, Turk and Pentland refer to these vectors as eigenfaces.
4
Limitations
The system effectiveness is highly dependant on the image
capture conditions.
•Variations in lighting conditions.
•Different lighting conditions for enrolment and query.
•Bright light causing image saturation.
•Differences in pose – Head orientation.
•2D feature distances appear to distort.
•Image quality.
•CCTV, Web-cams etc. are often not good enough.
•Expression (change in feature location and shape).
•Partial covering (Hats, scarves, glasses).
Meaning face recognition systems are usually not as accurate as
other biometrics.
5
Possible Solution
There are many image representations and filtering techniques
that can reduce the effect of lighting conditions and improve
image quality e.g.
•Colour normalisation
•Histogram equalisation.
•Edge detection.
•Noise reduction.
Could such methods as these be used to improve
eigenface based face recognition?
6
Test Database
960 bitmap images of 120 individuals (60 male, 60 female)
extracted from the AR Face Database provided by Martinez
and Benavente [10]. All images are translated, rotated and
scaled, such that the centres of the eyes are aligned.
The database is separated into two disjoint sets:
•The training set, (60 different people, natural lighting
conditions, neutral expression).
•The test set, (900 images - 15 images of 60 people,
captured under a variety of conditions).
After a series of initial investigations, we select a resolution of 25 pixels between
the eyes, with a width and height of 75 and 112 pixels respectively.
6300 comparisons are made to calculate false rejection rates and 7080
comparisons to calculate false acceptance rates.
7
Test Procedure
8
Output
FAR
The percentage of distance measures
below the threshold, when images of
different people are being compared.
FRR
The percentage of distance measures
above the threshold when images of
the same person are being compared.
By varying the threshold we obtain error rate pairs describing a curve.
The EER is used to compare pre-processing techniques.
However, it should not be used as a guideline to the system performance in a real
9
world situation.
Image
Filters
Convolution
Methods
Colour
Representation
Methods
Statistical
Methods
Method
Combinations
10
Results
11
Tweaking
A few final adjustments are made
to determine the optimum image
region. Reducing the ERR by
another 1.6%.
Original
Processed
Average
The first five eigenfaces generated from
the pre-processed training set.
12
Conclusion
•The eigenface-based method of face recognition can be significantly
improved by introducing a simple image pre-processing step.
•An EER of 22% percent can be achieved (a reduction of 12%) on a set of
extremely difficult images (20% partially obscured, 40% taken under extreme
lighting conditions).
•The remaining 22% error shows that some factors are not compensated for
by the single image pre-processing techniques.
•The effect of some filters may be mutually exclusive and could therefore be
used in parallel for improved results (further experimentation required).
•Other variations of PCA based face recognition (Pentland et al’s modular
eigenface system [8], Belhumeur et al’s comparison to Fisher faces [9]) have
also shown significant improvements, without image pre-processing. It is
likely that such systems could also benefit from the filtering techniques
described, resulting in greatly improved face recognition systems.
13
References
[1] M. Turk, A. Pentland. Eigenfaces for Recognition. Vision and Modelling Group, Massachusetts Institute
of Technology.
[2] M. Turk, A. Pentland. Face Recognition Using Eigenfaces. Vision and Modelling Group, Massachusetts
Institute of Technology.
[3] G.D. Finlayson, B. Schiele, J. L. Crowley. Comprehensive Colour Image Normalisation. The Colour and
Imaging Institute, The University of Derby.
[4] G. Finlayson, G. Schaefer. Hue that is Invariant to Brightness and Gamma. School of Information
Systems, University of East Anglia.
[5] Y. Adini, Y. Moses, S. Ullman. Face Recognition: the Problem of Compensating for Changes in
Illumination Direction. Department of Applied Mathematics and Computer Science, The Weizmann Institute
of Science.
[6] W. Zhao, R. Chellappa. 3D Model Enhanced Face Recognition. Sarnoff Corporation, Princeton.
[7] R. Cutler. Face Recognition Using Infrared Images and Eigenfaces. Department of Computer Science,
University of Maryland.
[8] A. Pentland, B. Moghaddom, T. Starner. View-Based and Modular Eigenfaces for Face Recognition. The
Vision and Modelling Group, Massachusetts Institute of Technology.
[9] P. Belhumeur, J. Hespanha, D. Kriegman. Eigenfaces vs. Fisherfaces: Recognition Using Class Specific
Linear Projection. Department of Electrical Engineering, Yale University.
[10] A.M. Martinez, R. Benavente. The AR Face Database. CVC Technical Report #24, June 1998.
14