The palmprint recognition has become a focus in biological recognition and image processing fields. In this process, the features extraction (with particular attention to palmprint principal line extraction) is especially important. Although great progress has been made, the representation of palmprint is still an open issue. We proposed a simple, efficient, and accurate palmprint principal lines extraction method in six steps: normalization, median filtering, average filters along four prefixed directions, grayscale bottom-hat filtering, combination of bottom-hat filtering, binarization and post processing. Our current work aims to integrate this method in a biometric identification system, in which features are derived from palmprint principal lines and used in a template matching algorithm.

1. +
Vito Gentile, Alessandro Bruno, Marco La Cascia
{vito.gentile,alessadro.bruno15,marco.lacascia}@unipa.it
Dipartimento di Ingegneria Chimica, Gestionale, Informatica, Meccanica
Università degli Studi di Palermo
Italy

2. +
Introduction
! The representation of palmprint is still an
open issue
! Main topics of this talk:
! A simple, efficient, and accurate palmprint
principal lines extraction method (six steps
method)
! How we aim to use these features in our future
works

3. +
Palmprint based biometry
! Palmprint-based biometric technology generally is
composed of two steps: feature extraction and recognition
! Image features are extracted and collected as input vector
while the remaining step is to recognize or classify that
vector into the suitable class to identify people
! We focused our attention on first step, more precisely, on
principal palm lines extraction

4. +
Pipeline of our algorithm
! The method consists of six steps:
! 1) normalization,
! 2) median filtering;
! 3) average filters (along 0°, 45°, 90°, 135°
directions);
! 4) grayscale bottom-hat filtering;
! 5) combinations of bottom-hat operations;
! 6) binarization and post processing.

5. +
Sample Images
! The ROI (region of interest) image is the central part of the
hand palm.
! We select the blue channel which showed best results,
compared to other choices
! 128 x 128 spatial resolution and 256 gray levels
(8 bit)

6. +
ROI Images
All the highlighted principal lines are oriented along directions
between 15° and 165°

7. +1. Normalization
2. Median Filter
3. Fi Filters
4. Grayscale bottom-hat
operation
5. Combination of
bottom-hat operations
6. Normalization
7. Binarization
8. Post Processing

8. +
1. Normalization
! The normalization aims to minimize the differences among
the contrast of input Images.
Input Image Output Image
The contrast is stretched

9. +
2. Median Filter
! To clean up the image, we use a median filter instead of a low
pass filter.
! Why Median Filter?
! minor effect on detail smoothing with respect to a low pass one,
while still reducing the salt and pepper noise, removing spikes.

10. +
3. Fi Filters
! Four average filters along four directions (0°,45°,90°,135°)
that we call F0, F45, F90 and F135

11. +
3. Fi Filters
! ImgX is filtered along the direction at X degrees from the
horizontal.
Img0 Img45
Img90
Img135
• Median filter is not applied to all images;
• Only to Img0, that is filtered with F0;
• At (0°) direction, rarely there are significant lines to be highlited

12. +
3. Fi Filters
! OutX is the filtered ImgX along the direction at X degrees
from the horizontal.
Out0
Out90
Out45
Out135

13. +
4. Bottom-hat Operation
! Bottom hat filtering is tipically used for binary images;
(1)
(2)
Subtracting the original image from the closing of it (from
morphological viewpoint)
Si(j,k) represent the value at (j,k) of the structural
element used for each operation.

14. +
4. Bottom-hat filtering (BHF)
! The result of bottom-hat operation is a low key image
Out0 BHF0

15. +
! We need to combine bottom-hat operations and product a
single image, that sum up the results included in each image.
! To this end, we decided to compute the average of the four
images
BHF0 BHF45
BHF90 BHF135
5. Combining bottom-hat operations

16. +
! The result of combination of bottom-hat filterings is a low
key image (very noisy)
! To improve the binarization we apply the eq. (3):
(3)
M is the mean value of the whole image, and k is a scaling coefficient
5. Combining bottom-hat operations

17. +
6. Thresholding & Binarization
! Binarization can be summarized with the following formula.
! Simple threshold method for the binarization (Threshold
Value = M)
(4)
ImgBW(x,y)

18. +
7. Post processing
! Simple Post processing is used to remove all the white and
isolated spikes (and blocks) in the image:
! Non-linear filter removing white, isolated spikes;
! For each pixel ImgBW(x,y), consider a 5x5 matrix
centered on it.
! If the first and last columns and rows contains only black
pixels, set all pixels of the matrix to be black (0).
Binarized Image Post processed Image

19. +
Experimental Results
Binarized Image Post processed ImageInput Image

20. +
Experimental Results
! The dataset consists of 1000 images, taken from CASIA, COEP
and randomly from internet;
! the groundtruth consists of 1000 binary hand labeled palmprint
images
! Test Images are 128 x128 grayscale images.
! Output of the system is a binary image
! Comparisons with 3 state of the art methods

21. +
Experimental Results
! True Positives, False Positives, False Negatives are evaluated
with respect to the ground truth (binary hand labeled palmprint
images)
! The evaluation of the performance of the approach is done by
using statistical indices, such as Precision (P), Recall (R),
F-measure (F):
! MD is the binary version of the detected palmprint map, while
MR is the binary version of the reference hand labeled principal
lines (ground truth).
(5) (6) (7)

22. +
Experimental Results

23. +
Future Works:
palmprint-based identification
! We aim to integrate our algorithm in a recognizer system
1. Store users’ palmprint-based features in a database
2. Use a template matching algorithm for the identification
! Palm print identification seeks to answer the question “who is
this person?” by examining his or her palm print.
! User’s palm lines are matched against each ones stored in the
database and the most similar template is obtained as the
identification result

24. +
Future Works:
palmprint-based identification
! Template matching: which distance measure?
! Hausdorff distance: a non-linear operator which measures the
mismatch of the two sets.

25. +

26. +
Conclusions
! The method we proposed is simple, efficient and accurate
with respect to some state of the art method;
! Precision, Recall, F-measure values are encouraging enough
to extend the test to the full dataset;
! Other Future Works:
! improvement of post processing for a better accuracy

27. +
References
! Our method:
Bruno,A.,Carminetti,P.,Gentile,V.,La Cascia,M.,& Mancino,E.
(2014,October).Palmprint principal lines extraction.In
Biometric Measurements and Systems for Security and Medical
Applications (BIOMS) Proceedings,2014 IEEEWorkshop on
(pp.50-56).IEEE
! Template matching based on Hausdorff distance:
You,J.,Li,W.,& Zhang,D.(2002).Hierarchical palmprint
identification via multiple feature extraction.Pattern
recognition,35(4),847-859
! Palm line matching:
Wu,X.,Zhang,D.,&Wang,K.(2006).Palm line extraction and
matching for personal authentication.Systems,Man and
Cybernetics,Part A: Systems and Humans,IEEE Transactions on,
36(5),978-987.

28. +
Thanks for your attention