3. Binary Discriminant Analysis
• The rationale of the proposed method is to use a set
of n linear discriminant function to transform a real-
valued template into a n dimensional binary face
template.
4. Binary Discriminant Analysis
• The key issue is how to find the optimal linear
discriminant functions so that the discriminability of
the binary templates is maximized.
• To maximize the discriminability Between class
variance should be maximum and Within class
variance should be minimum.
• To minimize within class variance, we use perceptron
learning rule
• To maximize between class variance we make use of
BCH codes.
5. Perceptron Learning Rule
• The perceptron is employed to find optimal linear
discriminant function (LDFs) so that the output binary
templates are close to the corresponding target binary
template. Thus, the within-class variance of the binary
templates is minimized.
• If two classes C1 and C0 are linearly separable, a linear
discriminant function
g(x) = wTx + t
is constructed so that for a random sample in these two
classes,
g(x) > 0, if x Ɛ C1
g(x) ≤ 0, if x Ɛ C0
9. Feature Extraction
• DCT Coefficient Scanning Order
Currently we are using first approx 500 coefficients in zig
zag order but it can be reduced to less number by taking
large no of sample of same facial image varying its
Illumination, Pose, Expression and Lighting conditions.
10. • BCH Codes
In our project, we have used BCH [255,37,45]
code.
• Hashing Algorithm
We have used SHA 256 hash algorithm to
secure the confidential data(password).
11. • Pre-processing
We have taken 256 bit length BCH code and
SHA 256 takes input length 512 bit hence we
need to preprocess it. Append the bit '1' to
the message .Append k bits '0', where k is the
minimum number >= 0 such that the resulting
message length (modulo 512 in bits) is
448.Append length of message (without the
'1' bit or padding), in bits, as 64-bit big-endian
integer (this will make the entire post-
processed length a multiple of 512 bits)
15. Experimental Analysis
• Between Class Variance
The BCH coding scheme preserves the
minimum hamming distance between its each
code. For BCH (255, 37, 45), the minimum
hamming distance is (2x45+1)=91 and the
corresponding variance is (91)^2/4 which can
be considered as strong between class
variance and also helps to improve
performance of our biometric cryptosystems.
16. Experimental Analysis
• Within Class Variance
In our project we have taken more number of
images of each individual having different pose
and expression. As the no of image increases, the
correlation between them also increases which
leads to less variance. . Another importance of
having more images of each individual is
decrease in storage requirements since only few
DCT coefficients or low frequency components
need to be retained.
17. Experimental Analysis
• Revocability
In our project we have adopted the BCH coding
scheme as a target which is further stored in the
database after employing hashing technique such
as SHA-256 for security purpose. This coding
scheme is applied over the user’s password. So if
the password is changed in case of templates
compromisation, its BCH code will be changed
and their corresponding hash will also be
changed which leads to a different data in the
database to be stored
18. Experimental Analysis
• Performance
In our project, database contains not directly
images but secured binary representation of
such images. So we can conclude that the
binarization and their hash don’t affect the
actual performance of our biometric
cryptosystem.
19. Experimental Analysis
• Security
In our project, we have used SHA-256 to
accommodate the security of face template
which provides the security level over birthday
attack of 2^128 which is quite enough for any
moderate level application.
20. Experimental Analysis
• Diversity
Since SHA algorithm is noninvertible and the
weight and bias matrix doesn’t reveal any information,
the imposter has to make an attempt for both what the
password is and what the relevant face should look
like. Therefore the brute-force attack is nearly
impossible for 5 digit password (94^5) consisting 94
different alphanumeric key. Since it is a two stage
verification systems, it provides enough security and
strong diversity for moderate level application.
21. Future Scope
• In order to achieve higher level of security, more
number of biometric traits are required to
authenticate whether the user is genuine or
imposter. So multimodal biometric cryptosystem
can be implemented by the same proposed
method along with other biometric traits.
• For higher level of security ,BCH coding schemes
and corresponding hashing techniques can be
changed to more number of bits as per
application requirements and number of users
22. References
S. Prabhakar, S. Pankanti, and A. K. Jain, “Biometric
Recognition: Security and Privacy Concerns,” IEEE Security and
Privacy Magazine, Vol. 1, No. 2, pp. 33-42, March-April 2003.
Anil K Jain, Ajay Kumar, Biometrics on Next Generation: An
Overview.
Jain AK, Nandakumar K, and Nagar A (2008) Biometric
template security. EURASIP J Advances n Signal Processing,
Special issue on Biometrics.
A. K. Jain, K. Nandakumar and A. Nagar, "Biometric Template
Security", EURASIP Journal on Advances in Signal Processing,
January 2008.
Stelvio Cimato, Marco Gamassi, Vincenzo Piuri, Roberto Sassi
and Fabio Scotti, Privacy in Biometrics.
23. References
A. Teoh Beng Jin, D. Ngo Chek Ling, and A. Goh. Biohashing:
two factor authentication featuring fingerprint data and
tokenised random number. Pattern recognition, 37(11):2245–
2255, 2004.
Hossein Malekinezhad, Hossein Ebrahimpour-Komleh,
Protecting Biometric-based Authentication Systems against
Indirect Attacks.
Hossein Malekinezhad, Hossein Ebrahimpour-Komleh, Fractal
Technique for Face Recognition.
Y C Feng1, Pong C Yuen1and Anil K Jain, A Hybrid Approach for
Face Template Protection.
Andrew B.J. Teoha, Yip Wai Kuan b Sangyoun Lee a,
Cancellable biometrics and annotations on BioHash, Pattern
Recognition 41 (2008) 2034 – 2044.