Gradient based vein Extraction Algorithm for Biometrics System

International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
Gradient based vein Extraction Algorithm for
Biometrics System
A. Parthiban K, B. Palanisamy A M
PG Scholar, Department of Information Technology, Bannari Amman Institute of Technology,
[email protected]
PG Scholar, Department of Computer Science and Engg, Bannari Amman Institute of Technology,
[email protected]
In Present days, Authentication by means of biometrics systems is used for personal verifications. In spite
of having existing technology in biometrics such as recognizing the fingerprints, voice/face recognition
etc., the vein patterns can be used for the personal identification. Finger vein is a promising biometric
pattern for personal identification and authentication in terms of its security and convenience. Finger vein
has gained much attention among researchers to combine accuracy, universality and cost efficiency. We
propose a method of personal identification based on finger-vein patterns. An image of a finger captured
under infrared light contains not only the vein pattern but also irregular shading produced by the various
thicknesses of finger bones and muscles. The proposed method extracts the finger-vein pattern from the
unclear image by using gradient feature extraction algorithm and the template matching by Euclidean
distance algorithm. The better vein pattern algorithm has to be introduced to achieve the better Equal Error
Rate (EER) of 0.05% comparing to the existing vein pattern recognition algorithms.
Keywords: Equal Error Rate (EER), Personal Identification Numbers (PINs), False Rejection Rate
(FRR) and False Acceptance Rate (FAR).
Personal identification technology is used in a wide range of systems for purposes such as
part access control, logins for PCs, bank ATM systems, investigation, driver identification, ecommerce systems and many more. Biometric procedures for identifying personalities are
attracting attention because conservative techniques such as keys, passwords, and PIN numbers
carry the risks of being stolen, lost, or forgotten. There has been considerable research in
biometrics ( Allain et al,1991), ( Berke,2010) over the last two decades. The list of physiological
and behavioral biometric characteristics that has to date been developed and implemented is long
and includes the face ( Lee et al, 2008); ( Jain et al,2004), iris (Kilic et al,2011), ( Li et al,2011),
fingerprint ( Miura et al,2004), palm print ( Novianto et al, 2002), hand shape ( Miura et
al,2004), voice ( Peleg et al,1984), signature ( Song et al,2011), and gait ( Wang et al,2010).
Notwithstanding this great and increasing variety of biometrics, no biometric has yet been
developed that is perfectly reliable or secure. For example, fingerprints and palm prints are
usually frayed; voice, signatures, hand shapes, and iris images are easily forged; face recognition
can be made difficult by occlusions or face-lifts; and biometrics such as fingerprints, iris and face
recognition are susceptible to spoofing attacks (Wang et al , 2010), i.e., the biometric identifiers
can be copied and used to create artifacts that can deceive many currently available biometric
devices. The great challenge to biometrics is thus to improve recognition performance and be
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
maximally resistant to deceptive practices (Yang et al , 2009). To this end, many researchers
have sought to improve reliability and frustrate spoofs by developing biometrics that are highly
individuating; yet at the same time, highly effective and robust. Finger vein pattern is just a
promising qualified candidate for biometric-based personal identification.
We designed a special device for acquiring high quality finger-vein images and propose a
DSP based embedded platform to implement the finger-vein recognition system in the present
study to achieve better recognition performance and reduce computational cost.
The rest of this paper is organized as follows. An overview of the system which proposed
here is in Section 2. The device for capturing the finger-vein image acquisition is introduced in
Section 3. Our recognition method is addressed in Section 4. Experimental outcomes are
discussed in Section 5. Finally, conclusion and future enhancement of the algorithm is described
in Section 6.
Overview of the System
The proposed system consists of three hardware modules: image acquisition module,
DSP main board, and machine communication module. The hardware diagram of the system is
shown in Figure. 1. The image acquisition module is used to collect finger-vein images. The DSP
main board including the DSP chip, memory (flash), and communication port is used to execute
the finger-vein recognition algorithm and communicate with the peripheral device. The human
machine communication module (LED or keyboard) is used to display recognition results and
receive inputs from users. A special imaging device is used to obtain the infrared image of the
finger. An infrared light irradiates the back side of the hand and the light passes through the
finger. A camera located in the palm side of the hand captures this light. The intensity of light
from the LED is adjusted according to the brightness of the image.
Figure.1. The hardware diagram of the proposed system.
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
Figure.2. The flow-chart of the proposed recognition algorithm
The proposed finger-vein recognition algorithm contains two stages: the enrollment stage
and the validation stage. Both stages start with finger-vein image preprocessing, which contain
detection of the region of interest (ROI), image segmentation, align the scanned image, and
enhancement. For the enrollment phase, after the pre-processing and the Gradient extraction
stage, the finger-vein template recorded into database is built. For the verification stage, the
finger-vein image is inputted for matched with the corresponding template after its features are
extracted. Figure 2 shows the flow chart of the suggested algorithm. Some altered methods may
have been proposed for finger-vein identical. In view of the computation complexity,
competence, and feasibility, however, we propose a novel method based on the Gradient theory,
which will be introduced in Section 4 in detail.
Image Acquisition
To obtain high quality near-infrared (NIR) images, a distinct device was developed for
acquiring the images of the finger vein without being affected by ambient temperature. Mostly,
finger-vein patterns can be imaged based on the principles of light reflection or light
transmission ( Novianto et al, 2002). We developed a finger-vein imaging device based on light
transmission for more discrete imaging. Our device mainly contains the following modules: a
monochromatic camera of resolution 580 × 600 pixels, daylight cut-off filters (lights with the
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
wavelength less than800 nm are cut off), transparent acryl (thickness is 10 mm), and the NIR
light foundation. The construction of this device is illustrated in Figure. 3. The transparent acryl
serves as the platform for locating the finger and removing uneven illumination. The NIR light
exposes the backside of the finger. In (Miura et al,2004), a light-emitting diode (LED) was
charity as the illumination source for NIR light. With the LED enlightenment source, however,
the shadow of the finger-vein perceptibly appears in the captured images. To address this
problematic, an NIR laser diode (LD) was used in our system. Compared with LED, LD has
robust penetrability and higher power. In our device, the wavelength of LD is 808nm. Figure 4
shows an example raw finger-vein image captured by using our device.
Figure. 3. Illustration of the imaging device.
Figure. 4. An example raw finger-vein image captured by our device.
Proposed Algorithm
A. Image Preprocessing
The Captured finger vein image can contain various noise and distortion on it. To extract
the feature vein patterns, the image captured have to be normalized by means of image preprocessing techniques. The resultant image is the high contrast image which is to be further
processed for the extraction of vein patterns by the algorithm proposed. The procedure for preprocessing the image as follows:
• Read the initial image
• Convert the RGB image to the Gray scale image
• Increase the contrast of the gray scale image by multiplying the image pixel value with
the constant
• Noise removal of the contrasted image by adding the “Salt and Pepper” noise onto the
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
• Distortion of the image has been done by using median filter
• Convert the image to Double precision image
The flow of pre-processing is shown below.
Figure. 5. Pre-Processing of Finger Vein Image
B. Image Enhancement
The pre-processed image has to further enhance to improve the contrast of the image. The
mean of the Double precision image has been identified and the contrast has been increased by
means of floating point accuracy (eps) of the image. The fig.6 shows the resultant contrast
images are shown below based on the power of eps.
Figure.6. Enhanced Image
C. Gradient Image Extraction
The feature veins have been extracted from the enhanced image by means of the
proposed Gradient Feature Selection Algorithm. An image of a finger captured under infrared
light contains not only the vein pattern but also irregular shading produced by the various
thicknesses of the finger bones and muscles. The gradient direction representation provides
better discrimination ability than the image intensity, and it shows that the combination of
gradient directionality and intensity outperforms the gradient feature alone.
In the proposed algorithm, the gradient magnitude has been identified by the given equation
| G| =
The Gx and Gy give the value of the n-dimensional filtering of the gray-scale image with
the sobel operator matrices in “Replicate” as border options. The Sobel operator creates the own
filter and performs a 2-D spatial gradient measurement of the image that corresponds to the
edges. The Absolute gradient magnitude at each point of the Input gray scale image is calculated
by the equation (1). The gradient is high at borders of the image and low at rest of the image as
shown in the fig 7.
The Gray scale image is converted to the double precision image for finding the gradient
magnitude. The double precision image along with the filter created by the sobel operator has
been used to perform the magnitude calculation.
For Gx, The Sobel matrix is given as
January.2014-Volume 2.No1
ISSN 2345-3397
International journal of Computer Science & Network Solutions
−1 0 1
Sx = −2 0 2
−1 0 1
For Gy, The sobel matrix is the transpose of Sx,
Sy=SxT (3)
−1 −2 −1
0 (4)
The Sx and Sy along with the Double precision grayscale image is processed separately
foe Gx and Gy respectively. Based on the equation (1), The Gradient Magnitude of the gray scale
image has been calculated. Figure 4.4 shows the resultant feature extracted image.
Sy =
Figure. 7. Gradient Feature Image
D. Feature Matching
Euclidean distance is considered from the center of the source cell to the center of each of
the neighboring cells. True Euclidean distance is calculated in each of the distance tools.
Conceptually, the Euclidean algorithm works as follows: for each cell, the distance to each
source cell is determined by calculating the hypotenuse with x_max and y_max as the other two
legs of the triangle. This scheming derives the true Euclidean distance, somewhat than the cell
distance. The shortest distance to a source is determined, and if it is less than the specified
maximum distance, the value is allocated to the cell location on the output raster.
The output values for the Euclidean distance raster are floating-point distance values. If
the cell is at an equal distance to two or more sources, the cell is assigned to the source that is
first encountered in the scanning process. You cannot control this scanning process.
The above description is only a conceptual depiction of how values are derived. The
actual algorithm computes the information using a two-scan sequential process. This process
makes the speed of the tool independent from the number of source cells, the distribution of the
source cells, and the maximum distance specified. The only factor that influences the speed with
which the tool executes is the size of the raster. The computation time is linearly comparative to
the number of cells in the Analysis window.
Experimental Results
A. Dataset for the experiment
The Dataset has been archived form different organizations. In the dataset that had taken
for i=the processing contains a set of 2110 finger samples. Each sample having the dimension of
170 x 76 Grayscale image. The dataset contains 106 sets where each set emphasis the 36 finger
image of different dimensions. From the archiving data of finger-vein based personal
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
authentication system introduced in section 2, we eliminate some not active users from the all
1000 users because they have too few records. Because of the vacancy of common finger-vein
image database for finger-vein recognition, we build an image database which contains 4500
finger-vein images from 100 individuals. Each individual contributes 45 finger-vein images from
three different fingers: forefinger, middle finger and ring finger (10 images per finger) of the right
hand. All images are captured using a homemade image acquisition system. The captured fingervein images are 8-bit gray images with a resolution of 320×240.
Figure. 8. Finger-vein images from different fingers after preprocessing
B. Performance Evaluation
a. False Acceptance Rate (FAR)
FAR is also called False Match Rate (FMR). It mentions to the prospect that the system
erroneously matches the input decoration to a non-matching template in the database in other
words, it procedures the percentage of inacceptable inputs which are erroneously accepted.
b. False Rejection Rate (FRR)
FRR is also called false non-match rate (FNMR). It is definite as the likelihood that the
organization fails to detect a match flanked by the input pattern and an identical template in the
database. That is, it procedures the percentage of valid participations which are imperfectly
rejected. It is judicious that the FAR reductions but the FRR increases due to the compassion of
the biometric device upsurges. In practical applications, the FAR should be very low to provide
high enough confidence and the FRR must be sufficiently low. If the threshold set in the decision
stage is reduced, it is expected that less false non-matches but more false accepts generated. In
other words, a higher threshold corresponds to a smaller FAR and a larger FRR.
c. Receiver Operating Characteristic (ROC) Curve
The ROC curve is used for illustrating the relationship between FAR and FRR. It is a
pictorial classification of the trade-off flanked by the FAR and the FRR, i.e., in a ROC curvature
the vertical and horizontal axes are FAR and FRR or vice versa, separately.
d. Equal Error Rate (EER)
EER is also called Crossover Error Rate (CER). It refers to the error rate at which the
FAR equals to the FRR and hence can be effortlessly gained from the ROC curve. In addition, it
is usually used for comparing the accuracy of devices with different ROC curves.
e. Failure to Enroll Rate (FER)
Besides FAR, FRR and ROC, there are other two factors usually considered in a vein
recognition system. One is the failure to enroll rate often caused by low quality inputs. It
resources the rate at which challenges to create a template from a participation is unsuccessful.
The other is called failure to capture rate. It refers to the probability that the system fails to detect
a correctly presented biometric input.
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
f. Response Time
In practical application, the response time must be taken into account. It is jointly
determined by two factors. Unique is the computational complication of vein recognition
algorithm and the extra is the competence of processing platform including the accepted
software, performance of CPU and the scope of memory, etc.
g. Performance for Personal Authentication
To examine the performance of proposed method for personal authentication, we did an
experiment using the method described in to evaluate the false accepted rate (FAR) and false
rejected rate (FRR). The proposed Algorithm has been related with the existing line tracking
method then the mean curvature method for Equal Error Rate (EER). Here by comparing all the
algorithms using the pattern normalization have lower error rates than the version without
Figure. 9. Genuine Accept Rate (GAR) at Different Thresholds
Figure. 10. Genuine Reject Rate (GRR) at Different Thresholds
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
Figure. 11. False Accept Rate (FAR) at Different Thresholds
Figure. 12. False Reject Rate (FRR) at Different Thresholds
Figure. 13. FRR (Type I Error) versus FAR (Type II Error)
C. Comparison with Previous Methods
Miura et al. y yn used a database that contained 1356 different infrared images of fingers.
These images were achieved from persons working in their laboratory aged 20 to 40,
approximately 70% of whom were male. Song’s finger-vein image dataset contained 5000
images together using an infrared imaging device they constructed. Seven images were taken for
each of 105 fingers. Compared with these databases, ours is greater and the data-collection
interval is longer. Thus, our database is more stimulating. Moreover, our system is implemented
on a general DSP chip. Table 1 shows that the average times required for feature extraction and
matching in our system are 225 ms and 12 ms, respectively. For the whole system, plus the time
for image capturing, the time required for the authentication of a user is less than 0.8 s. Even
though the feature extraction in our system is a little bit more difficult than that in Song's
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
method, our system achieves an EER of 0.05%, indicating that our method significantly
outperforms previous methods.
Feature Matching
in ms
in ms
Our method
Liu and Song’s Method
Miura’s method
0.05 225
0.07 343
0.145 450
Conclusion And Future Enhancements
In this paper, we introduced a Gradient Feature detector to extract vein patterns. It can
obtain all the points on the Gradient of vein in the image and increase the information of the
feature. We also proposed a new pattern normalization method, which can reduce the irregular
distortions caused by variance of finger pose. By using this method, we not only use the mutual
information among different vein branches, but also treat every vein branch with independence.
This is good for distinguishing the detailed differences among different finger vein patterns, and
helpful for dealing with the non-rigid deformation of portions of vein branches during the feature
matching process. The proposed system includes a device for capturing finger-vein images and a
proposed algorithm to extract finger-vein images by considering various parameters like vein
width, position, length, pixels and intersection of veins. Our system is suitable for mobile devices
and ATM’s because of its low computational complexity and low power consumption. The
advantage of this proposed system is more secured and confidential. The EER of 0.05% is
achieved which shows the better performance than the existing vein recognition algorithms.
In the future, this project can be extended to Hand vein, due to ease of access with more
security. For the hand vein we need to develop the high end system with several features to be
extended. The performance of the hand vein system has to be compared here. The Proposed
method extracts the vein features and save it as a template for matching or authentication. As a
feature enhancement, the extracted features have been combined with encryption or other
security technologies without diminishing authentication performance when converted by means
of encryption technology, the vein pattern has been encrypted to generate multiple feature codes
in a binary format from a single piece of biometric data, in the case of data leaks or theft; a new
feature code can be generated. This technique can be applied to fingerprint authentication as well
as palm vein authentication.
Allain and M. Cloitre, “Characterizing the lacunarity of random and deterministic fractal
sets”, Physical Review A, vol.44, no.6, pp. 3552- 558, 1991.
International journal of Computer Science & Network Solutions
January.2014-Volume 2.No1
ISSN 2345-3397
Berke, “Using Spectral Fractal Dimension in Image Classification”, innovations and
advances in computer sciences and engineering, pp. 37-241, 2010.
H. Lee, S. Lee, T. Kim, and Hyokyung Bahn, “Secure user identification for consumer
electronics devices,” IEEE Transactions on Consumer Electronics, vol.54, no.4, pp.17981802, Nov. 2008.
K.Jain, S. Pankanti, S. Prabhakar, H. Lin, and A. Ross, “Biometrics: a grand challenge”,
Proceedings of the 17th International Conference on Pattern Recognition (ICPR), vol. 2,
pp. 935-942, 2004.
K. I. Kilic and R. H. Abiyev, “Exploiting the synergy between fractal dimension and
lacunarity for improved texture recognition”, Signal pocessing, vol. 91, no. 10, pp. 23322344, 2011.
Li, H.B.; Yu, C.B.; Zhang, D.M. Study on finger vein image enhancement based on
ridgelet transformation. J. Chongqing Univ. Posts Telecommun. Nat. Sci. Ed. 2011, 23,
Miura, N. et al.: Feature Extraction of Finger-Vein Patterns Based on Repeated Line
Tracking and its Application to Personal Identification. Machine Vision and
Applications, Vol. 15, 2004, pp. 194–203.
Novianto, Suzuki, and Maeda, “Optimum estimation of local fractal dimension based on
the blanket method,” Transactions of the information Processing Society of Japan, vol.
43, no.3, pp. 825-828,2002.
N. Miura, A. Nagasaka, and T. Miyatake, “Feature extraction of fingervein patterns based
on repeated line tracking and its application to personal identification”, Machine Vision
Application, vol. 15, no.4, pp.194–203, 2004.
S. Peleg and J. Naor, “Multiple resolution texture analysis and classification IEEE
Transactions on Pattern Analysis and Machine intelligence, vol.6, no.4, pp.518-523,
W. Song, T. Kim, H. C. Kim, J. H. Choi, H. Kong and S. Lee, “A finger-vein verification
system using mean curvature”, Pattern Recognition Letters, vol. 32, no.11, pp. 15411547, 2011.
Wang , J. Li, and G. Memik, “User identification based on fingervein patterns for
consumer electronics devices”, IEEE Transactions on Consumer Electronics, vol. 56, no.
2, pp. 799-804, 2010.
Wang, K.J.; Liu, J.Y.; Popoola Oluwatoyin, P.; Feng, W.X. Finger Vein Identification
Based on 2-D Gabor Filter. In Proceedings of the 2nd International Conference on
Industrial Mechatronics and Automation, Wuhan, China, 30–31 May 2010; pp. 10–13.
Yang, J.F.; Shi, Y.H.; Yang, J.L.; Jiang, L.H. A Novel Finger-Vein Recognition Method
with Feature Combination. In Proceedings of the 16th IEEE International Conference on
Image Processing, Cairo, Egypt, 7–10 November 2009; pp. 2709–2712.