A Filterbank-Based Representation For Classification and Matching of Fingerprints

A Filterbank-based
Representation for
Classification and Matching of
Fingerprints
http://matlab-recognition-code.com
Figure 1 Description of the algorithm
Step
Corresponding M-function
Find the center (core point localization)
Supercore.m
Crop
Cropping.m
Sectorization
Whichsector.m
Normalize
Sector_norm.m
Filter-bank (8 filters)
Gabor2d_sub.m
Feature Vector of Variances
Sector_norm.m
Choice/Reject or add to DataBase
Fprec.m (main file)
Up to now, when a new fingerprint image is added to database, the FingerCode

was calculated two times: one time for input image and a second time for the
image rotated of a proper angle (22.5/2 degrees) in order to make the process
rotation-invariant (see the cited reference for more details). The image was
rotated using the Matlab function imrotate. This procedure can introduce noise.
To avoid this behavior we calculated the FingerCode associated to the rotated
image in this way: we rotate sectorization and the orientation of Gabor filters of
filter-bank of the same angle (22.5/2 degree). This is equivalent to consider as
filter-bank input a rotated image.
When a new fingerprint image is added to database, only one core point is found.
On the other side, when an input image is selected for fingerprint matching, a list
of candidates for core point is found and the matching is performed for each of
them. At last only the candidate with the smallest distance is considered. For
example , in database I have 3 images Img1, Img2 and Img3. Each of them is
characterized only by one core point, so I will have 3 core points, each of them
associated to an image present in database. If I select an image for fingerprint
matching (let it be ImgNew) I found for it a certain number of core point (let it N).
For each of these N core points (candidates) I will find the nearest fingerprint
image present in database. At last I will obtain N distances (as the number of
core points candidates): I say that the recognized image is the image with the
nearest distance I have obtained (this distance is associated to one of the initial
N core points candidates of ImgNew). This approach is very similar to the
algorithm discussed in
Erian Bezhani, Dequn Sun, Jean-Luc Nagel, and Sergio
Carrato, "Optimized filterbank fingerprint recognition",
Proc. SPIE Intern. Symp. Electronic Imaging 2003, 20-24
Jan. 2003, Santa Clara, California.
The
pixel-wise
orientation
field
estimation
(the
M-function
is
orientation_image_luigiopt.m) is greatly accelerated reusing previous sum
computations. The sum of elements of a block centered at pixel (I,J) can be used
for the computation of the sum of block elements centered at pixel (I,J+1). This
can be performed in the following way:
Once the sum of values centered at pixel (I, J) has been calculated (sum of
yellow pixels and orange pixels of Figure 1), in order to calculate the sum
centered at pixel (I, J+1) we simply subtract from the previous sum the yellow
area and add the green area (see Figure 2): in this way it is possible to save a lot
of computation. In other words
SUM (I, J) = yellow + orange
SUM (I, J+1) = SUM (I, J) yellow + green.
For more details concerning the implementation of this algorithm please visit
http://www.ece.cmu.edu/~ee551/Old_projects/projects/s99_19/finalreport.html
The function Supercore.m is discussed in detail in the document corepoint.doc
References
A. K. Jain, S. Prabhakar, and S. Pankanti, "A Filterbank-based Representation
for Classification and Matching of Fingerprints", International Joint Conference on
Neural Networks (IJCNN), pp. 3284-3285, Washington DC, July 10-16, 1999.
http://www.cse.msu.edu/~prabhaka/publications.html
An improved scheme for core point positioning

Step 1
Given an input image, I enhance [1] the fingerprint in order to obtain a better
image quality.
Figure 1. Input fingerprint image
Figure 2. Enhanced fingerprint image

Step 2
After this operation, the image is segmented and background is separated from
fingerprint image. This can be performed using a simple block-wise variance
approach, since background is usually characterized by a small variance. Image
is first binary closed (Matlab command imclose), then eroded (Matlab command
imerode), in order to avoid holes in fingerprint image and also undesired effect at
the boundary (between fingerprint and background). The image segmentation is
repeated several times, up to a desired condition is satisfied. This is done in
order to avoid undesired boundary effects between fingerprint and background.
The condition which has to be satisfied is chosen in the following way: the
enhanced image is divided into non-overlapping blocks of given sizes (usually 32
x 32 or 64 x 64). The whole enhanced image is filtered with a complex filter. Let
Cf_max be the maximum value of the filtered image in the current region of
interest (previously calculated according to some initial parameters). For each
non-overlapping block we calculate the relative maximum Cf_rel. We finally
consider a logical matrix F whose element (I,J) is equal to 1 if (I,J) is a block
relative maximum and this value is equal or greater than a threshold value
(usually 0.65*Cf_max in our simulations); F(I,J) is equal to 0 in all the other cases
(i.e. if F(I,J) is not a block relative maximum or a block relative maximum smaller
than the threshold value). If the number of non-zero elements of the logical
matrix F is above a threshold value, the parameters for image segmentation are
re-calculated and the whole process is repeated once again (complex filtering
output does not vary, only the region of interest has been changed).
Note: these relative maxima larger than the threshold value are not considered
core points but they will be considered candidates for it when a new image is
selected for fingerprint matching.
Figure 3. Image is segmented, closed and eroded (binary erosion and closing).
Step 3
Then we perform a fast pixel-wise orientation field computation [2].
Figure 4. Pixel-wise orientation field estimation in the region of interest.

Step 4
The orientation field is used to obtain a logical matrix where pixel (I,J) is set to 1 if
the angle of the orientation is <= PI/2 (PI 3.1415926535897...).
Figure 5. Logical matrix with local orientation field <= PI/2

Step 5
After this computation I find the border of this logical matrix, in the region of
interest of fingerprint image.
Figure 6. Border of the logical matrix in Fig.5 in the region of interest.

Step 6
After this computation I calculate the complex filtering output [3] of the enhanced
fingerprint image. It is not necessary to re-calculate it, I use the complex filtered
image calculated in step 2.
Figure 7. Complex filtering output of the enhanced fingerprint image.

Now I can find the maximum value of complex filtering output where the pixels of
the logical image (shown in Fig. 6) are set to 1.
Step 7
I repeat steps 4-5-6 for a wide set of angles (,PI/2-3*alfa, PI/2-2*alfa, PI/21*alfa, PI/2, PI/2+1*alfa, PI/2+2*alfa, PI/2 +3*alfa, where alfa is an arbitrary
angle). Each time I determine a point ( Note: each of them will be a candidate
for fingerprint matching).
Step 8
I subdivide all the points found in step 7 into subsets of
points which are quite near each other. I will have say N
subsets. For each subsets I will have a certain number of
candidates and I consider only subset with a number of
candidates >= 3. For each of this subset I consider the
subset with the greatest x-averaged coordinate. In this
subset I consider the core point the candidate with the
greatest x-coordinate. This is a good approximation in
standard fingerprint image.
See supercore7.m for more details concerning the algorithm used. The
functions supercore7_list returns as candidates for core point:
The points found at the end of step 8

The point found as relative maxima of non-overlapping block at the end of
step 2
Figure 8. Input fingerprint image and corresponding core point.
References
[1] S. Chikkerur,C. Wu and Govindaraju, "A Systematic approach for feature
extraction in fingerprint images", ICBA 2004
[2] Ravishankar Rao, "A taxonomy of texture description", Springer Verlag
[3] Kenneth Nilsson and Josef Bigun, Localization of corresponding points in
fingerprints by complex filtering, Pattern Recognition Letters 24 (2003), 21352144

A Filterbank-Based Representation For Classification and Matching of Fingerprints

Uploaded by

Copyright:

Available Formats

A Filterbank-Based Representation For Classification and Matching of Fingerprints

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Filterbank-Based Representation For Classification and Matching of Fingerprints

Uploaded by

Copyright:

Available Formats

A Filterbank-based

Figure 1 Description of the algorithm

Find the center (core point localization)

Feature Vector of Variances

Choice/Reject or add to DataBase

Fprec.m (main file)

Up to now, when a new fingerprint image is added to database, the FingerCode

The function Supercore.m is discussed in detail in the document corepoint.doc

An improved scheme for core point positioning

Figure 1. Input fingerprint image

Figure 2. Enhanced fingerprint image

Figure 4. Pixel-wise orientation field estimation in the region of interest.

Figure 5. Logical matrix with local orientation field <= PI/2

Figure 6. Border of the logical matrix in Fig.5 in the region of interest.

Figure 7. Complex filtering output of the enhanced fingerprint image.

The points found at the end of step 8

Figure 8. Input fingerprint image and corresponding core point.

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.