Face Recognition Using Simplified Fuzzy Artmap

Signal & Image Processing : An International Journal(SIPIJ) Vol.1, No.2, December 2010
DOI : 10.5121/sipij.2010.1212 134
FACE RECOGNITION USING SIMPLIFIED FUZZY
ARTMAP
Antu Annam Thomas1
and M. Wilscy2
1
Saintgits College of Engineering, Pathamuttom, Kottayam, Kerala
antuannam@gmail.com
2
Department of Computer Science, University of Kerala, Karyavattom Campus,
Trivandrum, Kerala
wilsyphilipose@gmail.com
ABSTRACT
Face recognition has become one of the most active research areas of pattern recognition since the early
1990s. This project thesis proposes a novel face recognition method based on Simplified Fuzzy ARTMAP
(SFAM). For extracting features to be used for classification, combination of Principal Component
Analysis (PCA) and Linear Discriminant Analysis (LDA) is used. This is for improving the capability of
LDA and PCA when used alone.PCA reduces the dimensionality of input face images while LDA extracts
the features that help the classifier to classify the input face images. The classifier employed was SFAM.
Experiment is conducted on ORL, Yale and Indian Face Database and results demonstrate SFAM’s
efficiency as a recognizer. The training time of SFAM is negligible. SFAM has the added advantage that
the network is adaptive, that is, during testing phase if the network comes across a new face that it is not
trained for; the network identifies this to be a new face and also learns this new face. Thus SFAM can be
used in applications where database needs to be updated frequently. SFAM thus proves itself to be an
efficient recognizer when a speedy, accurate and adaptive Face Recognition System is required.
KEYWORDS
Face recognition, Principal Component Analysis, Linear Discriminant Analysis, Neural Network,
Simplified Fuzzy ARTMAP.
1. INTRODUCTION
Face recognition [1] is an important and a challenging research problem spanning numerous
fields and disciplines. Face recognition is attention seeking because, in addition to having
numerous practical applications such as bankcard identification, access control, security
monitoring, and surveillance system, is a fundamental human behaviour that is essential for
effective communications and interactions among people.
Two categories of methods can be employed for face recognition one is global approach or
appearance-based approach and second one is feature-based or component-based approach.
Among these two categories of solutions to the problem [2] the most successful seems to be
appearance-based approaches [3], which generally operate directly on images or appearances of
face objects and process the image as two dimensional patterns. These methods extract features
to optimally represent faces belonging to a class and to separate faces from different classes.
Ideally, it is desirable to use only features having high separability power while ignoring the rest
[3].Most of the effort have been used to develop powerful methods for feature extraction [4]-[8]
and to employ classifiers like Support Vector Machine (SVM) [9], Hidden Markov Models
(HMMs) [10], Neural Networks [11]-[15] for efficient classification.
The main trend in feature extraction has been representing the data in a lower dimensional
space. Principal Component Analysis (PCA) [16],[17], [5]-[6] and Linear Discriminant analysis
(LDA) [7] are two main techniques used for data reduction and feature extraction in the
appearance-based approaches. PCA maximizes the total scatter while LDA maximizes the

135
between class scatter and minimizes the within class scatter. Thus extraction of features by PCA
is such that it is good for dimensionality reduction and data representation tasks while extraction
of features by LDA is such that it is good for recognition tasks [18]. But when dimensionality of
face images is high, LDA is not applicable and therefore LDA is deprived from its advantage to
find effective features for class separability. To resolve this problem PCA and LDA methods are
combined [3], PCA is applied to preprocessed face images to get low dimensionality images
which are ready for applying LDA.
Finally to decrease the error rate, instead of Euclidean distance criteria which was used in [19],
we implement a neural network, specifically SFAM, to classify face images based on its
computed LDA features
Earlier many of the face recognition systems used Back Propagation Neural Network (BPNN)
as the classifier. But BPNN suffers from Stability Plasticity Dilemma and takes a lot of
computer time to converge during the time of training [28]. Not only does SFAM overcomes
both these disadvantages but also turns the face recognition system to be adaptive.
The system was tested on ORL database and the test results demonstrated the efficiency
of SFAM as a recognizer with high recognition rate and negligible training time. Test
results also proved this novel system based on SFAM to be adaptive.
2. FACE RECOGNITION PROBLEM
In the literatures, face recognition problem can be formulated as: given an input face image
identify or verify the person in the image. When comparing person verification to face
recognition, there are several aspects which differ.
First, during verification, a client – an authorized user of a personal identification system –
makes an identity claim. Computationally this means that it is not necessary to consult the
complete set of database images in order to verify a claim. An incoming image (referred to as a
probe image) is thus compared to a small number of model images of the person whose identity
is claimed and not, as in the recognition scenario, with every image (or some descriptor of an
image) in a potentially large database. Second, an automatic authentication system must operate
in near-real time to be acceptable to users. Finally, in recognition experiments, only images of
people from the training database are presented to the system, whereas the case of an imposter
(most likely a previously unseen person) is of at most importance for authentication.
Face recognition is a biometric approach that employs automated methods to verify or recognize
the identity of a living person based on his/her physiological characteristics. In general, a
biometric identification system makes use of either physiological characteristics (such as a
fingerprint, iris pattern, or face) or behaviour patterns (such as hand-writing, voice, or key-
stroke pattern) to identify a person. Because of human inherent protectiveness of his/her eyes,
some people are reluctant to use eye identification systems. Face recognition has the benefit of
being a passive, non intrusive system to verify personal identity in a “natural” and friendly way.
All face recognition algorithms consist of two major parts: (a) face detection and normalization
and (b) face identification. Algorithms that consist of both parts are referred to as fully
automatic algorithms and those that consist of only the second part are called partially automatic
algorithms. This paper presents a partially automatic face recognition system.
3. FACE RECOGNITION PROCESS
A successful face recognition system should have a feature extractor that is be able to extract
features that help the classifier to efficiently classify the input face images. Feature Extraction

136
process is thus as important as classification process in every Face Recognition Algorithms.
Large amount of effort in object recognition field is thus put into developing methods by which
features can be extracted so that the objects are optimally represented and objects of different
class are sufficiently separated. Principal Component Analysis (PCA) and Linear Discriminant
Analysis (LDA) are two popular tools used for dimensionality reduction and feature extraction
in most of pattern recognition applications.
Principal component analysis or karhunen-loeve transformation [20] is standard technique used
in statistical pattern recognition and signal processing for data reduction and Feature extraction
[21]. As the pattern often contains redundant information, mapping it to a feature vector can get
rid of this redundancy and yet preserve most of the intrinsic information content of the pattern.
These extracted features have great role in distinguishing input patterns.
Thus PCA extracts features that ensure compact way of representation. But the problem is that
PCA not only maximizes the inter class scatter but also the intra class scatter.
Figure 1. Between-class (Inter-class) and Within-class (Intra-class) scatter
Figure 1 shows the Inter- class and Intra-class scatter. The within class scatter represents how
face images are distributed closely within classes and between class scatter describes how
classes are separated from each other. Thus to effectively recognize faces features should be
extracted so that Inter-class scatter should be maximized and Intra-class scatter should be
minimized. But in case of PCA it maximizes both Inter-class and Intra-class scatter.
So we go for a feature extractor that extracts features such that Inter-class scatter is maximized
and Intra-class scatter is minimized. LDA is such a feature extractor where Inter-class scatter is
maximized and Intra-class scatter is minimized. Thus LDA seeks directions that are efficient for
discrimination.
Figure 2(a) shows PCA (Eigen-faces) maximizing projected total scatter and Figure 2(b) shows
LDA (Fisher-faces) maximizing the projected interclass scatter.
Figure 2(a). Projection of sample
points onto a Eigenface space
Figure 2(b). Projection of sample
points onto a Fisherface space

137
Thus from this discussion it is clear that PCA is of great advantage when dimensionality
reduced optimal representation is required but it does not perform better than LDA in
recognition tasks [17],[22]. LDA is very efficient in selecting features that are of at most
importance in recognition tasks because LDA aims at increasing the inter-class scatter and
decreasing the intra-class scatter.
But LDA is disadvantageous when the dimensionality of face images is high. In order to resolve
this problem PCA and LDA is combined together so that PCA reduces the dimensionality of the
face images before LDA extracts features for recognition purpose.
The process of face recognition thus consists of four main steps Preprocessing, Dimensionality
Reduction, Feature Extraction and Classification.
The aim of the face preprocessing step is to normalize the face images, so that a robust feature
extraction can be achieved.
The aim of dimensionality reduction is to reduce the dimensionality of the preprocessed face
images. PCA is used in this step for dimensionality reduction.
The Feature Extraction step extracts features with high class separability. LDA is used in this
step for feature extraction.
The aim of Classification step is to classify the input face image to the correct class to which it
belongs.
Euclidean Distance approach is generally used for classification. But this method is error prone.
Thus we go for Neural networks which, with its massive parallelism in structure and high
computation rates, provide a great alternative to other conventional classifiers and decision
making systems.
Back Propagation Neural Network (BPNN) is generally employed as the classifier in Face
Recognition tasks. But BPNN suffers from Stability Plasticity Dilemma and High Computer
time for training. Thus in this paper a novel approach based on Simplified Fuzzy ARTMAP is
investigated.
Simplified Fuzzy ARTMAP overcomes all these disadvantages of BPNN. SFAM transformed
the system into an adaptive, speedy and more secure system.
3. SFAM THE CLASSIFIER
Simplified Fuzzy ARTMAP is the structure developed by Kasuba in the year 1993. It is
a vast simplification of Carpenter and Grossberg’s fuzzy ARTMAP. This has a reduced
computational overhead and architectural redundancy when compared to its predecessor
fuzzy ARTMAP.
SFAM comprises of two layers an input and an output layer (see Figure 3.) [23]-[26]. The input
to the network flows through the complement coder. Here the input string is stretched to double
the size by adding its complement also. The complement coded input then flows into the input
layer and remains there. Weights (W) from each of the output category nodes flows down to the
input layer. The category layer just holds the names of the M number of categories that the
network has to learn. The match tracking and vigilance parameter of the network architecture
are primarily for network training.
ρ, the vigilance parameter ranges from 0 to 1 and it controls the granularity of the output node
encoding. Thus, high vigilance values make the output node much fussier during pattern
encoding and low vigilance allows relaxed matching criteria for the output node.

138
The “match tracking” allows the network to adjust its vigilance during learning from some base
level, in response to errors in classification during the training phase. It is through match
tracking that the network adjusts its own learning parameter to decide when to create new
output nodes or reshape its decision regions. During training, match tracking is evoked when the
selected output node does not represent the same output category corresponding to the input
vectors given [19].
Figure 3. Architecture of SFAM network
Once SFAM has been trained, the equivalent of a “feed forward” pass for an unknown pattern
classification consists of passing the input pattern through the complement coder and into the
input layer. The output node activation function is evaluated and the winner is the one with the
highest value. The category of the input pattern is the one with which the winning output node
is associated [23]-[26].
4. THE PROPOSED SYSTEM
Figure 4 shows the proposed recognition system using SFAM as the classifier. Figure 4(a)
shows the training phase and Figure 4(b) shows the testing phase.
Figure 4(a). Training Phase of the Proposed Face Recognition System

139
Figure 4(b). Testing Phase of the Proposed Face Recognition System
The system consists of four steps: Preprocessing, Dimensionality reduction using PCA, Feature
extraction using LDA and Classification using SFAM.
The detailed explanation and implementation details of each step are given in the next session.
4. IMPLEMENTATION
Face recognition method is applied on ORL face dataset for separation of twenty classes. Out of
10 face images of each person 7 are used for training the network. Thus when we consider 20
classes there are 140 training face images. Figure 5 shows examples of sample images used.
Figure 5. Examples of Sample Face images used
Figure 6 gives the diagrammatic representation of step by step process followed in the
proposed system.

140
Figure 6. Steps of Implementation
Detailed explanation of each step in the system is given below.
4.1 Preprocessing
Figure 7. Flowchart of Preprocessing Step
In this phase first we resize the images to 40*40 sizes. After that we histogram equalize all input
images in order to spread energy of all pixels inside the image. As a next step, we subtract mean
images from face images to mean center all of them. Finally all preprocessed face images
change to a vector (1600 ×1vector) and go to the next step. Figure 8 shows example of
preprocessed images and Figure 7 shows the flowchart of preprocessing step.
Figure 8. Examples of preprocessed images
4.2 Dimensionality Reduction using PCA
Figure 9. Flowchart for Dimensionality Reduction step
Figure 9 gives the flowchart for Dimensionality Reduction step. Every input image is cropped to
40×40 image in the preprocessing step; as a result the input of this stage is a preprocessed
1600×1 vector. These vectors are used to estimate the covariance matrix. After estimation of the
covariance matrix, significant eigenvectors of the covariance matrix are computed. Number of

141
eigen-vector depends on the accuracy that the application demands. Large number of eigen-
vectors will obviously improve the accuracy of the method but computational complexity will
increase in this step and next step. Thus considering accuracy and computational complexity
100 eigen vectors are selected as principal eigen vectors.
The preprocessed faces are now projected on to the space with 100 eigenfaces as base vectors to
get 140, 100×1 vectors. Thus dimensionality reduction phase converts 1600×1 vectors to 100×1
vectors.
4.3 Feature Extraction using LDA
Figure 10. Flowchart for Feature Extraction using LDA
Figure 10 shows the flowchart of Feature Extraction step. Outputs of dimension reduction part
are 100×1 vectors. Features are extracted from these vectors. 19 principal components which are
selected from among the eigen vectors form the features. The preprocessed face images
projected on to the eigen space, is now, projected onto the fisher space and thus results in 140,
19×1 vectors which are used as input of neural network.
4.4 Classification using SFAM
For the direct application of SFAM to the face recognition problem, the architecture is trained
using the 140, 19×1 vectors got from the feature extraction stage. The vigilance parameter is
selected so that the number of categories to which the network settles is same as the number of
classes in the training data.
The network took only 2 epochs for completing its learning or training phase. The time that it
took for learning is negligible (order of seconds) and thus can be labeled as very efficient in
terms of time complexity.
Once the training is over and the top-down weight matrices have been obtained, the testing can
be carried out using the matrix equivalent of the test input image.
The most important point to note about the network in the classification stage is that it remains
open to adaptation in the event of new information being applied to the network. If an unknown
pattern is applied to the network SFAM will always attempt to assign a new class in the
category output layer by assigning the unknown input to a node.
Classification using BPNN was also implemented for comparison purpose. Figure 11 shows the
architecture of BPNN used for classification. The network implemented consists of an input
layer with 19 neurons, two hidden layers first with 30 and second with 25 neurons respectively
and an output layer with 10 neurons. A simple back propagation algorithm is used to train the
network.
Figure 11. Architecture of BPNN implemented

142
5. RESULTS
Recognition rate was observed by varying number of training images and testing
images. Experiment was conducted on ORL Database.
5.1 SFAM proved to be a speedy and accurate recognizer
Table 1. Result of the Experiment Conducted on ORL Database
No: of Training
Images Used for
Each Subject
No: of
Testing
Images
Used for
Each
Subject
Name of the
Classifier
No :of Epochs
Elapsed Time
(in seconds)
Recognition
rate
7 3
BPNN 100 1255.03 95.5%
SFAM 2 0.125 98.5%
6 4
BPNN 100 1250 94%
SFAM 2 0.115 98%
5 5
BPNN 100 1253 93%
SFAM 2 0.120 97%
Table 1 shows the result of the test conducted on ORL database. Three cases (number of
training samples taken as 7,6 & 5) are considered here.
The recognition rate for system using SFAM as classifier is found to be higher than the
recognition rate for system using BPNN.
The training time for SFAM is negligible when compared to BPNN. The training time for
BPNN is about 10,000 times that of training time for SFAM. This adds to the advantage of
the system using SFAM as the classifier.

143
Figure 12. Graph showing the Result of experiment conducted on ORL Database
Figure 12 shows the plot of recognition rate versus number of training samples, when
the test is carried out on ORL Database. The graph shows that the recognition rate for
the system using SFAM as the classifier is higher when compared to the system using
BPNN as the classifier.
The above result analysis show that SFAM exhibits a very good performance both in
terms of recognition rate and time required to train the net. The recognition rate can be
improved by increasing the number of eigen vectors selected as Principal Components
in the dimensionality reduction step.
The network took only 2 epochs for completing its learning or training phase. The time
that it took for learning is negligible (order of seconds) and thus can be labeled as very
efficient in terms of computer time. Thus SFAM proves itself to be a classifier that
learns very fast when compared to Back Propagation Neural Network that takes more
than 1200 seconds to converge when trained for the same input.
Thus recognition system using SFAM can perform as an efficient recognizer in cases
where a speedy and accurate system is required.
5.1 SFAM proved to be adaptive
When the face recognition system using BPNN as the classifier is given any input image that
belongs to a class for which system is not trained for, it will converge to the conclusion that it
belongs to one of the class for which it is trained.
An example is shown in Figure 13. Figure 13(a) shows the test input belonging to the class for
which the system is not trained for.

144
Figure 13(a) Test Image belonging to the class, Figure 13(b) Classification that BPNN
network is not trained for yields
Figure 13(b) shows the class to which the system using BPNN as the classifier converges.
But in the case of SFAM it conclude that the input does not belong to any class for which it
trained for by giving the output as classification= -1 , then the network learns this input as a
pattern belonging to a new class.
After this learning procedure if the network come across this input again it classifies it to the
new class it created.
Thus from this result it is obvious that the system using SFAM as the classifier is open to
adaptation in the event of new information being applied to the network. If the system using
SFAM as the classifier encounters a face image that it is not trained for, it will conclude that it
is a face that it is seeing for the first time and it learns the new pattern. But in the case of system
using BPNN as the classifier erroneous input (test input not belonging to any of the class for
which it is trained for) yields erroneous result.
Since SFAM can adaptively be trained, this can be employed in the cases, like student record
maintenance at institutions, attendance keeping at organizations and so on, where addition to
database needs to be made frequently.
6. CONCLUSION
In this work a new, SFAM based Face recognition, method is presented. Proposed method
consists of four parts: i) Image Preprocessing that includes histogram equalization, and mean
centering, ii) Dimensionality reduction using PCA where main features that are important for
representing face images are extracted iii) Feature Extraction using LDA where significant
features for class separability are selected and iv) SFAM classifier classifies input face images
into one of available classes if the test input belongs to a class for which it is trained for and if
not the system concludes that it is an input belonging to a class it is not trained for and will learn
this new pattern. Back Propagation Neural Network is also implemented in this work for the
comparison purpose. When compared to BPNN, SFAM acts as a more powerful classifier by
exhibiting negligible computer time for training and a solution to Stability Plasticity Dilemma.
Simulation results using ORL, YALE and Indian Face Database demonstrated the ability of
SFAM over BPNN for efficient face classification. Not only did SFAM made the face
recognition system adaptive and produced a high recognition rate but also reduced the training
time by around 10,000 times when compared to BPNN. Thus this work proves system using
SFAM as the classifier to be a speedy (negligible training time), secure ( high recognition rate)
and adaptive face recognition system.Future works include enriching the system to attain more
human-like recognition capabilities and consequently provide tools that are more targeted to
user’s actual needs. This target in the future could be met by making the system successful even
in vastly varying pose and lighting conditions. This is possible only when feature extractors are
extracting features such that this type of improved recognition is possible. So powerful feature
extractors can be employed to extract features strong enough to make the recognition possible in
much less constrained situations. This work has proved SFAM (as a classifier) to be successful
both in terms of recognition rate and training time required, for the given set of features, when

145
compared to other classifiers. So the features extracted with more powerful feature extractors
along with SFAM as the classifier will turn out the system to be more human like.
REFERENCES
1. Jain, R. Bolle, S. Pankanti Eds, “BIOMETRIC – Personal Identification in Networked
Society”, Kluwer Academic Publishers, Boston/ Dordrecht/ London, 1999.
2. J. R. Solar, P. Navarreto, " Eigen space-based face recognition: a comparative study of
different approaches, IEEE Tran. , Systems man And Cybernetics- part c: Applications, Vol. 35,
No. 3, 2005.
3. Sahoolizadeh, H.; Ghassabeh, Y.A.;” Face recognition using eigen-faces, fisher-faces and
neural networks”, Cybernetic Intelligent Systems,2008. CIS 2008.7th IEEE International
Conference on 9-10 Sept. 2008,Page(s):1 - 6 Digital Object Identifier 10.1109 / UKRICIS.
2008.4798953
4. M. Turk, A. Pentland, "Eigen faces for face recognition", Journal cognitive neuroscience,
Vol. 3, No.1, 1991
5. W. Zhao, R. Chellappa, A, Krishnaswamy, “Discriminant analysis of principal component
for face recognition”, .IEEE Trans. Pattern Anal. Machine Intel., Vol 8, 1997.
6. O.Deniz, M. Castrill_on, M. Hern_andez, “Face recognition using independent component
analysis and support vector machines” , Pattern Recognition letters, Vol. 24, PP. 2153-2157,
2003.
7. John Wright, Allen Y.Yang, Arvind Ganesh, S. Shankar Sastry and Yi Ma, “Robust Face
Recognition via Sparse Representation,” IEEE Trans. On Pattern Analysis and Machine
Intelligence, vol. 31, no. 2, pp. 210-225, 2008
8. Thang V.Pham and Arnold W.M. Smeulders, “Sparse Representation for Fine and Coarse
Object Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28,
no. 4, pp. 555-567,2006.
9. K. Lee, Y. Chung, H. Byun, "SVM based face verification with feature set of small size",
electronic letters, Vol. 38, No. 15, PP. 787-789, 2002.
10. H. Othman, T. Aboulnasr, " A separable low complexity 2D HMM with application to face
recognition" IEEE Trans. Pattern. Anal. Machie Inell., Vol. 25, No. 10, PP. 1229-1238, 2003.
11. M. Er, S. Wu, J. Lu, L.H.Toh, "face recognition with radial basis function(RBF) neural
networks", IEEE Trans. Neural Networks, Vol. 13, No. 3, pp. 697-710.
12. M.J.Er, W.Chen, S.Wu, “High speed face recognition based on discrete cosine transform
and RBF neural network”, IEEE Trans. On Neural Network, Vol. 16, No. 3, PP. 679,691, 2005.
13. Z. Pan, A. G. Rust, H. Bolouri,, “ Image redundancy reduction for neural network
classification using discrete cosine transform”, Proc. Int. Conf. on Neural Network, Vol. 3, PP.
149,154, Italy, 2000.
14. [14] Nazeer, Shahrin Azuan, Nazaruddin Omar, and Marzuki Khalid. "Face Recognition
System using Artificial Neural Networks Approach." International Conference on Signal
Processing, Communications and Networking. Chennai, India: IEEE, 2007. 420-425
15. Ming Gu Jing-Zhou Zhou Jian-Zhong Li “Online face recognition algorithm based on
fuzzy ART” ,International Conference on Machine Learning and Cybernetics, 12-15 July 2008,
Vol. 1, Page(s): 556 - 560
16. M.A. Turk, A.P. Pentland, “Face Recognition Using Eigenfaces”, Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition, 3-6 June 1991, Maui, Hawaii, USA,
pp. 586-591
17. D.L. Swets and J.J. Weng , “Using Discriminant Eigen features for image retrieval”, IEEE
Trans. Pattern Anal. Machine Intel, vol. 18, PP. 831-836, Aug. 1996.
18. A. M. Martinez, A. C. Kak, “PCA versus LDA”, IEEE Trans. Pattern Anal. Machine Intell,
Vol. 23, pp. 228-233. 2004.

146
19. P.N. Belhumeur, J.P. Hespanha, and D. J. Kriegman, “Eigen faces vs. Fisher faces:
Recognition using class specific linear projection”, IEEE Trans. Pattern Anal. Machine Intel.,
vol. 19, PP. 711-720, may 1997.
20. Papoulis, A. & Pillai, U. (2002).”Probability, random variables, and Stochastic Processes”,
McGraw-Hill, 0073660116, New York
21. Haykin, S. (1999) ” Neural Networks: A comprehensive foundation, Prentice Hal”l, 0-13-
273350-1,New Jersey
22. Eleyan, A. & Demirel, H. (2006). “PCA and LDA Based Face Recognition Using
Feedforward Neural Network Classifier”, Proceedings of Multimedia Content
Representation,Classification and Security, pp. 199-206, 978-3-540-39392-4, Turkey,
September 2006,Springer-Verlag, Istanbul
23. S.Rajasekaran and, G.A.Vijayalakshmi Pai,” Image Recognition Using Simplified Fuzzy
ARTMAP Augmented with a Moment based Feature Extractor”, International Journal of Pattern
Recognition and Artificial Intelligence, Vol. 14, No. 8 (2000) 1081–1095
24. S.Rajasekaran and, G.A.Vijayalakshmi Pai, “Neural Networks, Fuzzy Logic, and Genetic
Algorithms”, Prentice-Hall of India, New Delhi, 2003.
25. T. Kasuba, “Simplified fuzzy ARTMAP,” AI Expert, November (1993) 18–25.
26. S. Rajasekaran and V. Pai, “Simplified fuzzy ARTMAP as a pattern recognizer”, J. Comput.
Civil Engin. ASCE 14, 2 (2000) 92–99.
27. Kwang In Kim, Keechul Jung, and Hang Joon Kim, “Face Recognition Using Kernel
Principal Component Analysis”, IEEE Signal Processing Letters, February 2002, Vol. 9, No:2,
40-42
28. R.Beale and T.Jackson,” Neural Computing”,Institute of Physics publishing, Bristol and
Philadelphia, 2001.

Face Recognition Using Simplified Fuzzy Artmap

More Related Content

What's hot (20)

Similar to Face Recognition Using Simplified Fuzzy Artmap (20)

Recently uploaded (20)

Face Recognition Using Simplified Fuzzy Artmap