Академический Документы
Профессиональный Документы
Культура Документы
Abstract— Forensic Science is an art and science of a print made by an impression of ridges in the skin of a
finger, often used for biometric identification in criminal investigation. The law enforcement agencies uses system
like AFIS (Automatic Fingerprint Identification System) where reference fingerprints are stored in database
which are further used to match with latent fingerprints recovered from actual crime scene. There is high rate of
rejection of latent fingerprints since they are found in damaged condition due to blood spills, oil spills, wet
surface, snow, dust, etc that damages their quality and hence individuality of fingerprint is lost. Verification of
such mutilated fingerprints against stored reference fingerprint can be done using Laplacian, Gaussian, minutia-
based, etc methods. Each such method has there own advantages and disadvantages. The paper emphasis on the
use of Gabor filters for fingerprint identification. The fingerprint matching is done by extracting the Finger code
from both reference and latent fingerprints and then finding the Euclidean distance between the two
corresponding finger codes. After proper training the system, the result obtained provides 99 % rate of
recognition.
Input image
N −1 N −1
1
Is
VAR ( I ) = 2 ∑ ∑ ( I (i, j ) − M ( I )) 2 -- (2)
Match
N i=0 j=0
Found
? It is a short time Fourier transform analysis of given input
fingerprint image. Normalization is done in order to remove
the unwanted noise which get added in the input fingerprint
Match found Match found image due to sensors [2].Normalization is done on each sector
Euclidean Euclidean separately by considering its estimated mean and variance.
The input fingerprint image I (x, y) can be normalized by
Distance Distance =0
using equation:
=abc.xyz
(V0 * I ( x , y ) − M i ) 2
N (x, y ) = M
i 0 +
Vi
; ifI ( x , y ) ≥ M
-- (3)
(V0 * I (x , y ) − M i ) 2
Store the feature in the feature library N (x, y ) = M
i 0 −
Vi
; otherwise
-- (4)
2.6 Filtering
The sinusoidal-shaped waves of ridges and valleys vary
slowly in a local constant orientation which provides useful
information. As Gabor filter have both frequency selective
and orientation selective properties and provides optimal joint
resolution in both spatial and frequency domains therefore we
can use Gabor filter as band-pass filter that can be tuned to
the corresponding frequency and orientation and thus can
effectively remove the unwanted noise and helps to restore
the lost minutia points, ridge and valley structure. The even-
symmetric Gabor filter has the general form [6, 7] as:
1 X 2 Y 2
h ( x , y : Φ , f ) = exp − 2Φ + Φ2 cos (2πfx Φ ) -- (5)
2 δ x δ y
2.7 Finger code generation nearest match which will display Euclidian distance (say,
The Mean, Standard deviation, Co-occurance, i.e Contrast 996.9907, 1450.9073, etc.). Now the mean of all such
in each sector will represent the feature vector and are defined Euclidean distance of 8 fingerprint images are taken (e.g. for
as: 101_1.tif series fingerprint images: mean of Euclidean
distance is 813.2008).This is shown in Table 1. Note that, if
we resubmit any fingerprint from available template, say
1 k 101_4.tif, then the system will calculate the finger code of
Mean ( Mi θ ) = ∑ Si
k i =1
-- (7) fingerprint image 101_4.tif and will match it with all the
available finger codes and will come with output of Euclidean
distance equal to zero and with corresponding Image ID.This
indicates that the recognition rate achieved is 99 %.This is
Std .deviation ( Fi θ ) = (∑ k
)
Riθ ( x , y ) − Mi θ -- (8) shown in figure 5.
Suppose, if a new template of fingerprint image other than
above (say 101_9.tif), is provided as input then it is expected
that the value of Euclidean distance must be zero or lie in
n
between zero and less than or equal to the respective mean of
Contrast = ∑ (i −
i, j =0
j ) 2 C 2 (i, j ) -- (9)
Euclidean distance (say 0 ≤ (ED of 101_9) ≤ 813.2008).If
this is not the case, then it is clear that the provided input
fingerprint image does not match with the stored template.
The experimental results tabulated at serial no.2 up to serial
Where,
no.10 for different sets of fingerprint images shows that if the
calculated Euclidean distance of any image is found to be
I= 0, 1, 2, 3 …47.
greater than the respective mean of Euclidean distance then it
θ = 0°,45°,90°,135°,270°. is clear that it is not the perfect match but if it is less than the
K=Total number of pixels in the sector Si respective mean of Euclidean distance then it shows a perfect
Riθ = is the sector of filtered image match. In all such cases, it is highly expected from the user of
the system to add the finger code of every fingerprint image,
C= Co-occurrence matrix
irrespective of its match found or not found since this helps to
increase the stored template database and in turn matching
Here, the gray level value in each sector of the filtered
probability of the system.
image will give the finger code.