International Journal of Computer Applications (0975 – 888)

Volume 47– No.6, June 2012

Palm print Classification based on inter-Distal Region

Texture features

Dhananjay D.M C. V. Guru Rao, PhD. I. V. Muralikrishna, PhD.

Research Scholar Professor & Head Former Director, R & D Cell
JNTU Department of CSE JNTU
Hyderabad,India S.R.Engineering College Hyderabad,India
Warangal, India

ABSTRACT called as Tri-radiated section. Palm sample shown in Figure 2

Palm print is a widely accepted biometric trait for depicts different region or sections of palm print along with
authentication due to clarity in discriminating the features of principle lines. A close observation of tri-radiated section
palm such as large distance among non-class samples as well reveals that this region has good number of features to
as minimum distance between intra-class samples. Area differentiate between individual palm print samples. Some of
beneath the finger and enclosed by heart line is called as tri- the features that can be captured in tri-radiated section are
radiated region or inter-distal region. This specific area of ridges, whorl, arch, minutia lines etc. If we use tri-radiated
palm contains features which are unique and universally section as ROI to compare two samples, size of sample will be
discriminating. In this paper we present a simple method to limited at the same time we have good number of features to
extract ridges in tri-radiated section also called as inter-distal validate a given sample. In this paper we use tri-radiated
region of palm. Different orientation of ridges extracted from section to represent palm print sample and propose a
inter-distal region appears as a fine texture. We use these computationally effective method for validation of palm print
fine textures for validating samples of Palm print. Reduction sample.
in size of the image, it’s optimal storage, retrieval,
computational efficiency without compromising the fine
features of the palm sample and use of simple discriminating
features to validate a given palm sample has motivated this
paper. The results confirm the proposed methodology in this
paper is most efficient one.
ROI
General Terms
Pattern Recognition, Digital forensic. (Region Of
Keywords
Inter-distal region, Tri-radiated section, low pass filter, Interest)
Gabor filter, texture feature, Manhattan distance.

1. INTRODUCTION
Palm print based authentication is being used commercially
for establishing identity for secured access of resources [1].
When compared to finger print, palm print has not gained
wider acceptance due to the sample size. A solution widely Figure1. Region of Interest (ROI) in sample
followed to reduce the area of palm sample is to select a Palmprint (Courtesy: PolyUDatabase)
square region by using a universally accepted reference point.
One such reference point being used is parallel line drawn
below the base of fingers to extract region of interest (ROI)
[2] as shown in Figure 1. A submitted palm sample can be
interpreted as collection of different regions or section
demarcated based on the principle lines or major lines of the Heart Line
palm sample. Three principle lines that are prominent in every
palm print are Heart line, Head Line and Life line [3]. The
Head line
heart line normally originates from area below the small
finger and terminates below the index finger and is also
known as distal crease. Major regions which make 75% of the Fortune Line
palm print sample area are Thenar region, Hypothenar region
and Tri-radiated region. Terms inter distal region and tri- Life Line
radiated section or region are used interchangeably in this
paper. They are also referred as Region-I, Region-II and
Region III respectively. Region-I or Tri-radiated section is
area beneath the base of the finger enclosed by Heart line. The Figure 2. Palm print Regions and lines
area enclosed by life line is known as Thenar region. Area In this paper section 2 describes different features of tri-
opposite to Thenar region is known as Hypothenar region. radiated region. Pre-processing of sample is presented in
The area enclosed by this line below the base of finger is section 3. Section 4 explains the use of Gabor filter in six

11
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.6, June 2012

different orientations for generating sub samples. Simple Result of this operation is shown in fig (4a, 4b).
texture based feature extraction method is illustrated in
section 5. Section 6 analyzes the results obtained on the
proposed method. The research work presented in this paper is
concluded in Section 7 followed by the references.

2. FEATURES OF TRI-RADIATE REGION

Tri-radiated section occupies around 20% of the total area of
palm print. Area beneath the index finger up to little finger,
and below the heart line is designated as tri-radiated section or
region. The theory of palmistry regards this area as the one
containing maximum features than compared to any other
area. A close observation of this area reveals that many
features which can be studied with finger print are also
present in this area. While finger print normally works with
thumb sample as finger print, Ti-radiated section has multiple
such areas adjoining to each other. This feature can be useful
to increase the reliability of the palm print authentication Figure4a. Input sample Binarized
process. Ridge features, minutiae features, whorl formation of
ridge, arch formation, frequency of ridges in given area,
orientation of ridges in different area, ending point of ridges,
formation of island due to ridges and crossing of ridges are
some of the useful features that are useful parameters to be
used to discriminate one region from other region[4]. Another
advantage with this area is the listed features are rotation and
scale independent. Further the methods which are well
established in finger print authentication process can be
directly used here. As mentioned in earlier discussion the size
of sample will also be optimal hence enhancing the
authentication computational process. A sample area of tri-
radiated section / inter distal region captured using
conventional method is shown in fig (3).

Figure4b. Filtered Sample

4. SUB SAMPLE GENERATION USING

GABOR FILTER
Pre processed palm print image samples are applied to the
Gabor filter using six orientations to generate sub samples in
six directions. Response of ridges in specific direction is
consolidated and is used for feature extraction using texture
calculations. Gabor filter, Gabor filter bank and Gabor
wavelet are being used in feature extraction in image
processing due to their fine scale response controlled by
different orientation and scale [9]. A general Gabor
expression for spatial domain is as given eqn (3).

Figure3. Sample Inter-distal/Tri-radiated Section

( x 2  y 2 ) / 2
3. PRE-PROCESSING OF TRI-ADIATED G( x, y, u,  )  1 2 e e(uxisin uyi cos ) / 2 xy
REGION … (3)
For pre-processing of sample palm print we use a global
threshold value to binarize the image sample using eqn (1).
This process removes the background pixel [5]. We apply low
Where i =  1 , u is the central frequency of the
sinusoidal wave and in spatial domain controls the pixel
pass filter [6] to remove high frequency components using
resolution, θ controls the orientation of the function and σ is
eqn(2).
the standard deviation of the Gaussian envelope. We use θ in
1 if im(i, j )  T six orientations to capture the response of the filter. The six
IM   ... (1) directions used are (0, 30, 60, 90, 120, and 150) the other
0 directional responses are mirror image of the directional set
used. Following Figure 5 shows the process used for
IM  im (i, j )  LowPass Filter ... (2)
obtaining the filtered image in six directions further the
images are applied threshold to obtain the required response.

12
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.6, June 2012

Correlatio n   (i   x )( j   y ) pd (i, j )
i j

... (8)
For each direction of Ѳ degree Gabor filter response is
obtained. As we use six different values for Ѳ we get six
responses for each sample and calculate five texture
a. Input Sample b. Pre-processed sample features[11] as mentioned in eqn(4,5,6,7,8) given above. This
results in generating rich texture feature vector. The following
table (1) depicts the texture feature calculated for each
direction for one of the sample palmprint from data base.

Table1. Feature Vector calculated for sample-1

0 30 60 90 120 150
Direction
Feature

Energy 326380 152200. 156980. 360150. 214770. 160160.

c. 0 Degree d. 30 degree e. 60 degree
Entropy 99733.0 36399.0 38611.0 107840. 62456.0 39342.0

Homogeneit
y 2009.00 1446.00 1361.00 2081.00 1516.00 1422.00

Contrast 0.06 0.04 0.04 0.05 0.05 0.04

Correlation 2009.00 1446.00 1361.00 2081.00 1516.00 1422.00

f. 90 Degree g. 120 Degree h.150 degree 5.2 Palm Print Classification Based On
Figure 5. a) Input Sample b) Pre-processed sample
Texture Features.
Feature vectors obtained for each sample are used for finding
c) to h) Gabor filter response the class of palmprint. We have used Manhattan distance [12]
5. TEXTURE BASED FEATURE given in eqn(9) classifier for assigning proper class to
palmprint. If P1 is one point at coordinate (x1, y1) and P2 is
EXTRACTION AND PALMPRINT another point at (x2,y2) then the distance between two points
CLASSIFICATION P1,P2 is given by
5.1 Texture feature extraction.
Response obtained from Gabor filter in six directions depicts
Manhattn Dist  x1  x2  y1  y 2 ... (9)
a fine distribution of ridge orientation with respect to direction
Ѳ. This fine distribution of ridge response appears as a spatial For each sample and each directional response distance Di for
domain texture distribution. Texture can be defined as a i=1:6 is calculated and the final classification will be based on
regular repetition of an element or pattern in a given finding out the different samples to which the sample is
image[10]. Texture features are studied under statistical nearest. The maximum number of time a sample Si is closer to
domain and syntactic domain [7] [8]. In our paper we have submitted palmprint from database stored, that sample is
used five texture features Energy, Entropy, Contrast, obtained as the final class of the sample.
Homogeneity and Correlation defined by eqn(4,5,6,7,8)
respectively. 6. RESULTS AND DISCUSSION
Energy   pd (i, j ) 2
... (4) 6.1 Results
We have used samples collected using conventional inking
i j
method. In this method ink pad ink is spread on Inter Distal
region and the palm is pressed against a blank good quality
Entropy   pd (i, j ) log pd (i, j ) ... (5) paper. Three instances of same sample at different times are
i j used. The samples are scanned at 300 dpi. Format for the
image used is BMP. We have calculated for each sample
Contrast   (i  j ) pd (i, j )
based on texture feature which sample is nearest. Results of
... (6) this calculation are presented in table (2). If a Sample
i j
S i Max Occurance{ DBSamplei } then class of

Homogeneity   pd (i, j ) /(1 | i  j |) sample is DBSamplei , where DBSamplei sample from

i j database and Si is input test sample. Accuracy of our results
... (7) obtained by our proposed method is 99.33%. The results
obtained for few samples are presented in fig (6).

13
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.6, June 2012

Table 2. Samples classified as per texture features

Cor-
Homo-
Sample Entropy Energy Contrast geneity relation

1 9 9 12 9 8

2 4 4 3 4 9

3 5 5 2 5 4
(d)
Figure6. Test results for few samples from database
4 2 2 5 2 3

5 3 3 4 3 7 6.2 Discussion
Biometric systems reliability can be assessed by False
6 7 7 8 7 8 Rejection Rate which is major of response of a system
for valid input sample and number of attempts made by
7 6 6 9 6 5
person, This is represented in following graph fig(7). As
we change the threshold values to measure FRR, it can
8 6 6 6 6 6
be seen that when threshold is increased by 50% the
9 1 1 1 1 12
acceptance rate is 80.33 and the acceptance rate is nearly
60% when threshold is changed by nearly 200%. This
10 11 11 11 11 11 result proves the reliability of system of our proposed
method.
11 10 7 5 10 10

12 1 1 1 1 9

13 15 14 14 12 14

14 13 13 13 15 13

15 14 12 12 11 12

7. CONCLUSION
Enforcement of security and authentication is inevitable
these days which invariably employ the more effective
biometric systems with utmost computational efficiency
as well as precision and accuracy. Our proposed palm
(a) print based method reduces the computational burden.
Further the fine formation of ridge structure has best
discrimination features. The system ensures to capture
(b) good quality input samples. Texture features are good
measure for classification but do not imply a proper ridge
formation. The palm print based biometric system is
tested with many number of samples and found more
efficient than many existing other biometric systems.
Therefore it is highly recommended for personal
authentication.
(b)
8. REFERENCES
[1] A.K. Jain, R. Bolle, and S. Pankanti “Biometrics:
Personal Identification in Networked Society”, Kluwer
Academic, 1999.
[2] D. Zhang, A.W. Kong, J. You and M. Wong “Online
Palmprint Identification,” IEEE Trans. PAMI, vol. 25,
No. 9, Sept. 2003 PP. 1041-1050.
[3] H. V. Alexander ,“Classifying Palm Prints”, Charles C.
(c) Thomas Publication, Illinois, 1973 162 Pages.

14
 International Journal of Computer Applications (0975 – 888)
Volume 47– No.6, June 2012

[4] N. Duta, A.K. Jain, and K.V. Mardia, “Matching of [9] Hans G. Feichtinger, Thomas Strohmer,“Gabor Analysis
Palmprint”, Pattern Recognition Letters, vol. 23, no. 4, and Algorithms", Birkhäuser, 1998; ISBN 0817639594.
2001 PP 477-485.
[10] C. H. Chen, L. F. Pau,P. S. P. Wang “The Handbook of
[5] Gonzalez, Woods “Digital Image Processing” 3rd Pattern Recognition and Computer Vision”, 2nd Edition
Edition Eastern Economy Publication pp. 207-248, World Scientific Publishing Co., 1998.
[6] A.K Jain, Feng, J.J “Latent Palmprint Matching”, PAMI [11] Manesh Kokare, P. K. Biswas, and B. N. Chatterji
Vol. 31, No. 6, June 2009, PP. 1032-1047. “Rotation-Invariant Texture Image Retrieval Using
Rotated Complex Wavelet Filters”, IEEE Transactions
[7] Rafal Kozik and Michal Chora “Combined Shape and On Systems, Man and Cybernetics-Part B: Cybernetics,
Texture Information for Palmprint Biometrics” Journal Vol. 36, No. 6, Dec. 2006, PP 1273-1282.
of Information Assurance and Security vol.5 2010 PP 60-
66. [12] Richard O. Duda, Peter E. Hart “Pattern Classification”,
2nd Edition, Wiley Publication.
[8] Ghandehari, Azadeh, Safabakhsh,“Palmprint verification
Using Circular GaborFilter”, ICB09 PP 675-684.

15