Developing an Algorithm for 2D Face Recognition
Using Soft Computing
By
Ali Hussein Mohammed
Department of Information Technology Institute of Graduate
Studies and Research,
Alexandria University, Egypt
Introduction
• Facial recognition (or face recognition) is a type of
biometric software application that can identify a
specific individual in a digital image by analyzing
and comparing patterns. In face recognition
applications, original input image is of high
dimension. Feature extraction is the most important
step in face recognition, which reduces high
dimensional image data to low dimensional feature
vectors.
Authentication v.s Identification
 Face Authentication/Verification (1:1 matching)
 Face Identification/Recognition (1:N matching)
Applications of Face Recognition
Face Recognition
• The face recognition system operates in four-stages:
(1) Capture - a physical or behavioral sample is
captured by the system during enrollment; (2)
Extraction - unique data is extracted from the
sample and a template is created; (3) Comparison the template is then compared with a new sample
;(4) Matching - the system then decides if the
features extracted from the new sample are
matching or not.
Challenge Facing 2D Face Recognition
 Different persons may have very similar appearance
(Face recognition systems can't tell the difference
between identical twins).
Twins
Father and son
Challenge Facing 2D Face Recognition
• Faces with intra-subject variations in pose,
illumination, expression, accessories, color,
occlusions, and brightness.
Challenge: Representation
Age Invariance
 Facial shape and texture change over time.
 Applications
 Age specific access control
 Missing children, multiple enrollment
Why is Face Recognition Hard?
• Many Faces of the same person
Taxonomy of face Recognition
 face recognition methods can be categorized into two groups:
feature-based and appearance-based. In feature-based
approach, a set of local features is extracted from the image
such as eyes, nose, mouth etc. and they are used to classify the
face. The major benefit of this approach is its relative
robustness to variations in illumination, contrast, and small
amounts of out-of-plane rotation. But there is generally no
reliable and optimal method to extract an optimal set of
features. Another problem of this approach is that it may
cause some loss of useful information in the feature
extraction step.
Taxonomy of face Recognition
The appearance-based approaches use the entire
image as the pattern to be classified, thus using all
information available in the image. However, they
tend to be more sensitive to image variations. Thus
major issue of designing an appearance-based
approach is the extraction of useful information
which can be used for efficient face recognition
system that is robust under different constraints
(pose, illumination, expressions etc.)
Taxonomy of face Recognition
• Further classification of different approaches of face
recognition for still images can be categorized into
tree main groups such as holistic approach, featurebased approach, and hybrid approach. Face
recognition form a still image can have basic three
categories, such as holistic approach, feature-based
approach and hybrid approach.
Taxonomy of face Recognition
• Holistic matching methods. These methods use the whole
face region as the raw input to a recognition system .One
of the most widely used representations of the face region
is Eigen pictures, which are based on principal component
analysis.
• Feature-based (structural) matching methods. Typically, in
these methods, local features such as the eyes, nose, and
mouth are first extracted and their locations and local
statistics (geometric and/or appearance) are fed into a
structural classifier.
• Hybrid methods:- these methods used both feature-based
and holistic features to recognize a face. These methods
have the potential to offer better performance than
individual holistic or feature based method.
Taxonomy of face Recognition
Eigenfaces
• A set of eigenfaces can be generated by
performing a mathematical process called
principal component analysis (PCA) on a large
set of images depicting different human faces.
• Informally, eigenfaces can be considered a set of
"standardized face ingredients", derived from
(statistical analysis )(covariance matrix ) of many
pictures of faces. Any human face can be
considered to be a combination of these standard
faces.
Eigenfaces
• These Eigenfaces can now be used to represent both
existing and new faces: we can project a new
(mean-subtracted) image on the Eigenfaces and
thereby record how that new face differs from the
mean face.
• The eigenvalues associated with each Eigenface
represent how much the images in the training set
vary from the mean image in that direction.
• We lose information by projecting the image on a
subset of the eigenvectors, but we minimize this loss
by keeping those Eigenfaces with the largest
Eigenvalues.
Fisherfaces
• Similar to the Eigenface approach, yet able to account for
variations between multiple images of the same person.
• Utilises a larger training set containing multiple images of
each person.
• The ratio of between-class and within-class scatter matrices
is calculated.
• The eigenvectors of this matrix are then taken to formulate
the projection matrix.
• The low dimensional sub-space created maximises betweenclass scatter, while minimising within-class scatter.
Comparison
Eigenface
- Fast.
- easy implementation.
Fisherface
- Light invariant.
- better classification
Holistic Features
Soft Computing V.s Hard Computing
• Soft computing differs from conventional (hard)
computing in that, unlike hard computing, it is
tolerant of imprecision, uncertainty, partial
truth, and approximation. In effect, the role
model for Soft computing is the human mind.
• The principal constituents, i.e., tools, techniques,
of Soft Computing (SC) are – Fuzzy Logic (FL),
Neural Networks (NN), Support Vector Machines
(SVM), Evolutionary Computation (EC), and –
Machine Learning (ML) and Probabilistic
Reasoning (PR).
Soft Computing V.s Hard Computing
Soft Computing
Soft constraints
Hard Computing
real-time constraints
need of robustness rather need of accuracy and
than accuracy
precision in calculations and
outcomes
useful for routine tasks that useful in critical system
are not critical
Problem Statement
 Face recognition is a very difficult problem with
applications in many areas such as surveillance,
digital libraries, smart environments and security.
 Uncertainty is an intrinsic part of intelligent
systems used in face recognition applications.
The use of new methods for handling inaccurate
information about facial features is of fundamental
importance.
Problem Statement (Cont.)
Face recognition mainly depends on several different
factors:
 Facial expression such as sadness, happiness,
and facial pose.
 Occlusion: faces may be partially occluded by
other objects (like wearing glasses).
 Imaging conditions like lighting and camera
resolution.
 Presence or absence of structural constituents
like beards, mustaches and glasses.
Objective
 Reliable techniques for face recognition under
more extreme variations caused by pose,
expression, occlusion or illumination (highly
nonlinear) have proven elusive.
 A good face recognition methodology should
consider facial’s features representation as well as
classification issues.
Proposed System
 This thesis proposes deals with the design of
intelligent 2D face recognition system using interval
type-2 fuzzy logic for diminishing the effects of
uncertainty formed by variations in light direction,
face pose and facial expression.
 Built on top of the well-known fisherface method,
our system employs type-2 fuzzy set to compute
fuzzy within and in-between class scatter matrices
of fisher’s linear discriminant.
Proposed System (Cont.)
 This employment makes the system able to
improve face recognition rates as the results of
reducing the sensitivity to substantial variations
between face images.
 Type-2 Fuzzy Sets (T2FSs) have been shown to
manage uncertainty more effectively than Type-1
Fuzzy Sets (T1FS), because they provide us with
more parameters that can handle environments
where it is difficult to define an exact membership
function for a fuzzy set.
Methodology
Fig 1. Bock diagram
of the proposed face
recognition system
Methodology(Cont..)
 The system utilizes PCA as data representation to project
face patterns from a high-dimension image space to some
low dimensional space while retaining as much variation
as possible in the data set.
 Furthermore, it employs an enhanced approach for fuzzy
fisherface classification that helps us to find the optimal
classification–driven projections of face patterns that could
establish a high degree of similarity between samples of
the same class and a high degree of dissimilarity between
samples of many classes.
Methodology(Cont..)
 Feature Extraction Stage
 This stage relies on transformation of face samples by
utilizing PCA to derive a starting set of features.
 PCA is a well-known technique commonly exploited in
multivariate linear data analysis.
 The main underlying concept is to reduce the
dimensionality of a data set while retaining as much
variation as possible in the data set
Methodology(Cont..)
 Interval Type-2 Fuzzy K –Nearest Neighbor
 To improve the performance of the classifier, the
proposed system utilizes interval type-2 fuzzy K-NN
(IT2FKNN) to refinement of classification results so that
they could affect the within-class and between-class
scatter matrices.
 This stage assigns membership as a function of the
pattern distance from its K–nearest neighbor and those
neighbor's memberships in the possible classes.
Methodology(Cont..)
 FLD Classifier (Fisherface linear Discriminant)
 Taking into account the membership grades obtained from
IT2FKNN, the statistical properties of the patterns such as
the mean value and scatter covariance matrices are used
find the optimal classification–driven projection of patterns.
 Membership grades are incorporated into the definition of
the between-class scatter matrix and within-class scatter
matrix to get the fuzzy between-class scatter matrix and
fuzzy within-class scatter matrix
Simulations & Validation
 The algorithm is tested on YALE and ORL database to compute recognition rate.
The images of the same person are taken at different times, under lightly varying
lighting conditions and with various facial expressions. Some people are captured
with or without glasses. The heads in images are slightly titled or rotated.
 In our experiments, we split the whole database into two parts evenly. One part is
used for training and the other part is for testing. In order to make full use of the
available data and to evaluate the generalization power of algorithm more
accurately, we adopt across-validation strategy and run the system 10 times. In each
time, f face images from each person are randomly selected as training samples,
and the rest is for testing.
Simulations & Validation (Cont.)
The first experiment was performed using different images
per class for training, and the remaining images for testing.
For feature extraction, we used respectively PCA, LDA,
fuzzy Fisherface, fuzzy inverse FDA, 2DFLD and the
proposed system. Note that all methods involve a PCA
phase.

As we can see, the proposed interval type-2 fuzzy
fisherface outperformed other classification techniques.
Since other Fuzzy-based recognition methods may
preserve unwanted variations due to lighting and facial
expression, the recognition show a poor performance.
Comparison of Recognition Rate For
Yale Database
100
90
80
70
60
50
Case 1
Case 2
Eigenface(PCA)
Fuzzy fisherface(FDA)
Case3
Fisherface(LDA)
proposed system
Comparison of Recognition Rate For
ORL Database
100
95
90
85
80
75
70
65
60
55
50
Case 1
Eigenfaces (PCA)
Case 2
Fuzzy fisherface(FDA)
Case 3
Fisherface(LDA)
Proposed system
Table 1: Average Recognition Rate On The Face
Databases
Simulations & Validation (Cont.)
 In the second experiment, the recognition rate as performance index of different
face recognition algorithms are plotted with number of images per subject used for
training. From Fig. 2 we can see that the proposed system outperforms the other
methods for every number of training samples for each class. This is because, the
proposed system can extract more discriminative feature, in which using IT2FKNN
to get the membership degree matrix, FDA with the redefined fuzzy within-class
scatter matrix and fuzzy between-class scatter matrixes more efficiently captures
the distribution of samples than LDA and FDA.
Fig. 2 Recognition Rate for ORL
database
Conclusions
 The system calculates membership degree matrix through a generalized version of
fuzzy KNN algorithm called interval type-2 fuzzy KNN that includes refined
information about class membership of the patterns. By doing this, the system is
able to reduce sensitivity variations between face images caused by varying
illumination, viewing conditions and facial expressions.
 Unlike previous face recognition efforts based on fuzzy fisherface in which the
number of neighbors in KNN classifier is usually experiment-driven and needs to be
adjusted for a specific dataset at hand, our system is based on interval type-2 fuzzy
set to extend the membership values of each pattern as interval type 2 fuzzy
memberships by using several initial K in order to handle and mange uncertainty
that exist in choosing initial K.
 As a further study, we plan to examine generalized type-2 fuzzy sets such as zSlices
to improve system classification.