You are on page 1of 3

v

ABSTRACT
The fingerprint recognition system is considered to most important
biometric system in addition to other biometrics recognition systems. The
fingerprint recognition problem can be fingerprint verification and
fingerprint identification.
Fingerprint verification refers to authenticity of a person by his
fingerprint. The user provides fingerprint together with identity
information. In the verification process template is retrieved based on the
identification provided and matching is performed.

fingerprints based upon the unspecified conditions. In the identification of


fingerprint, the process matches fingerprints with the fingerprint database
for similarity.
A good fingerprint is required for the best verification and
identification tasks. Many approaches are available for fingerprint
verification and identification. One method is based on minutia,
representing the fingerprint by its local features, like terminations and
bifurcations. The other method is based on image processing. In this
method, matching is based on the features of the image.
This research work uses decomposition of fingerprint image by
using wavelet method. The wavelet type used is db-1, and coiflet. The
fingerprint image is decomposed to five levels. In each level of
decomposition, the fingerprint image is split into four parts namely:
approximation matrix, vertical matrix, horizontal matrix and a diagonal
matrix. The subsequent level of decomposition uses an approximation of
the previous level for further decomposition.
Statistical features are calculated at each level of decomposition
using all the 4 coefficient matrices. The statistical features are used as
inputs for training the artificial neural network (ANN) and Fuzzy logic
algorithms.
vi

The purpose of using ANN in the research work is because, existing


methods are working based on statistical parameters. The purpose of
using ANN for fingerprint recognition is due to the following reasons:
1. The working concepts of ANN are based on statistics like using
linear summation between layers to propagate information from
input to output layers,
2. ANN use transformation function like exponential function in
Radial basis function network, exponential function in back
propagation algorithm to squash output values from neurons,
3. ANN uses objective function for finding optimal weights between
layers for mapping inputs (statistical parameters) to outputs
(Labelling of fingerprints).
Because of the working properties of ANN are based on statistical
concepts, ANN assures improved fingerprint recognition.
ANN is the mathematical representation of the functioning of neural
connections in the human brain. The mathematical representation varies
depending upon the application. The ANN uses training algorithms. These
algorithms are grouped under supervised, unsupervised and recurrent
algorithms. The supervised algorithms use inputs and target outputs, the
unsupervised algorithm uses only inputs and recurrent algorithms use
inputs and target outputs along with previous outputs for training the
topology of the ANN.
The algorithms used for training the ANN topology are back
propagation algorithm (BPA), radial basis function (RBF). The features
used for training these ANN algorithms are the statistical features
obtained from wavelet coefficients.
The back propagation algorithm uses exponential function and
adopts repeated training process to learn the given input target patterns.
The BPA stops training once an error criteria is achieved.
vii

A multilayer feed forward network that uses an exponential


activation function is called radial basis function (RBF). The RBF uses the
concept of distance between patterns and the various centers of the
patterns. The number of nodes in the hidden layer of the RBF network is
equivalent to the number of centers used to find the distance. A bias
value of 1 is appended to the hidden layer nodes for convenience of
weights processing. The final weights are obtained between hidden layer
and the output layer. These weights are used for registering the images.
Fuzzy logic is implemented for recognition of the fingerprints. The
training features used for Fuzzy logic are the statistical features of the
wavelet coefficients.
The fingerprint images are considered form standard database.
Fingerprints are also generated. The BPA is trained with patterns of
statistical features obtained from Wavelet decomposition. Various
parameters
Performance of the algorithms are assessed by receiver operating
characteristic curve, sensitivity, specificity and accuracy.

The major contribution the thesis is as follows:


1. BPA is combined with wavelets to identify fingerprints.
2. RBF is combined with wavelets to identify fingerprints.
3. Fuzzy logic is combined with wavelets to identify fingerprints.
4. The accuracy of Fuzzy logic with wavelet in identifying
fingerprint is higher compared to that of BPA with wavelets
and RBF with wavelets.

You might also like