Palmprint verification using lagrangian decomposition and invariant interest
Palmprint Verification using Lagrangian
Decomposition and Invariant Interest Points
Authors: P. Gupta, A. Rattani, D. R. Kisku,
*C. J. Hwang, J. K. Sing
Presented by - C. J. Hwang
Department of Computer Science,
Texas State University,
San Marcos, Texas 78666, U.S.A
25 - 29 April 2011
Resort & Convention
Outline of Talk:
• Palmprint Biometrics
– Hand Geometry
– Palm Characteristics
• Advantages of Palmprint Trait
• Proposed Palmprint Identification System
• ROI Extraction
• Feature Extraction using SIFT
• Palmprint Matching using Graph
• Experimental Results
– CASIA Database
– IIT Kanpur Database
• Biometrics authentication is a method by which one can
be recognized based on one or more intrinsic physical or
behavioral human characteristics.
• Hand Geometry
– Features: Hand shape and dimensions, finger size
• Palm Characteristics
– Features: Principal lines, wrinkles and creases
• Principal lines: Heart line, head line and life line
• Wrinkles: Weaker and irregular lines, much thinner than
• Creases: More like fingerprint structure, have ridges and
Advantages of Palmprint System:
– High distinctiveness
– High permanence
– High performance
– Non – intrusiveness
– Low resolution imaging
– User – friendly
– Low price palmprint devices and low setup cost
– Highly stable
Proposed Palmprint Identification
• ROI  is detected and extracted from palm image.
• Uniform intensity distribution is obtained by applying
• SIFT is applied to the ROI (region of interest) of
palmprint image to extract invariant features
• Palmprint matching is performed using Lagrangian
decomposition and graph matching technique
To extract ROI of palm image the following steps are
• Convert the palm image to a binary image. Gaussian
smoothing is used to enhance the image.
• Apply boundary-tracking algorithm to obtain the
boundaries of the gaps between the fingers. Since the
ring and the middle fingers are not useful for processing.
Therefore, boundary of the gap between these two
fingers is not extracted.
• Determine palmprint coordinate system by computing the tangent of
the two gaps with any two points on these gaps. The Y-axis is
considered as the line which joining these two points. To determine
the origin of the coordinate system, midpoint of these two points are
taken through which a line is passing and the line is perpendicular to
• Finally, extract ROI for feature extraction which is the central part of
Feature Extraction using SIFT:
The scale invariant feature transform, called SIFT
descriptor, has been proposed by and proved to be
invariant to image rotation, scaling, partly illumination
changes and projective transform.
The basic idea of the SIFT descriptor is detecting feature
points efficiently through a staged filtering approach that
identifies stable points in the scale-space.
Local feature points are extracted from the following steps:
• Scale-space extrema detection: select candidates for
feature points by searching peaks in the scale-space from a
difference of Gaussian (DoG) function
• Keypoint localization: localize the feature points by using
the measurement of their stability
• Orientation assignment: assign orientations based on
local image properties
• Keypoint descriptor: calculate the feature descriptors
which represent local shape distortions and illumination
Palmprint Matching using Lagrangian
• Let G1 and G2 be two graphs obtained from a pair of palmprint images after
having extracted SIFT features
• A permutation matrix is determined from the pair of graphs and this
permutation matrix is used to minimize the distance between these graphs.
• Permutation matrix is a zero-one matrix whose rows and columns sum to
one. Rows and columns can add up to one or zero.
• In the deterministic annealing framework, permutation matrix constraints
can be formulated. The rows and columns constraints are known as winner-
• The proposed approach gets inspired by Lagrangian decomposition
approach in which the rows and columns constraints are satisfied
separately by Lagrange multipliers which are used to equate the two
• Let us consider, G1x1y1 and G2x2y2 are the two adjacency matrices of
two graphs G1(V,E) and G2(V,E), respectively. Now, a permutation
matrix M will be determined which will minimize the distance between
the two graphs.
• The adjacency matrices of the two undirected graphs can be
represented as symmetric and sparse matrices with zero diagonal
• The problem can be defined as follows
∑ ∑ ∑
xx y y
∑ ∑ ==
• The match matrix M contained in the distance measure must satisfy
the permutation matrix constraints. In addition to these constraints it
also includes row and column WTAs
∑ ∑ ==
• Let us consider two match matrices are given as and
which have the following properties,
∑ ∑ ==
• In the given constrains these properties are always satisfied. The
properties given in above equation can be established by taking a
new objective which is given by
• The constraint given in the above equation can be established using
a Lagrange parameter λ.
• Finally, the distance is compared with a predefined threshold and
accordingly decision of acceptance or rejection is made.
21 xxxx MM =
• CASIA Database
– 5502 palmprint images / 312 subjects
– Left and right palms
– 8-bit gray scale JPEG images
– Taken with uniform-colored background
– Uniform distributed illumination
– Normalized to 150×150 pixels
• IIT Kanpur Database
– 800 palmprint images / 400 subjects
– Resolution is set to 200 dpi
– Images are rotated by at most ±35 degree
– Images are normalized to 150×150 pixels
Table 1. FRR, FAR and Recognition Rates Determined on CASIA
and IIT Kanpur Databases
FAR (%)FRR (%)DATABASE
• In this paper, a palmprint based verification system using SIFT
features and Lagrangian network graph technique has been
• Region of interest (ROI) has been extracted from the wide palm
texture at the preprocessing stage and histogram equalization
technique is applied to palmprint image for obtaining uniform
• At the next stage, SIFT feature extraction is performed on palmprint
image, whereas only the ROI is considered for invariant points
• Finally, identity is established by finding permutation matrix for a pair
of reference and probe palm graphs drawn on extracted SIFT
features. Permutation matrix is used to minimize the distance
between two graphs.
• The experimental results computed on CASIA and IITK palmprint
databases show the effectiveness and the robustness of the
 Zhang, L., and Zhang, D., “Characterization of palmprints by wavelet signatures via directional context
modeling,” IEEE Transactions on Systems, Man and Cybernetics – B, 34(3), 1335 – 1347 (2004).
 Zhang, D., Kong, W. K., You, J., and Wong, M., “On-line palmprint identification,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, 25, 1041 – 1050 (2003).
 Han, C. C., Cheng, H. L., Lin, C. L., and Fan, K. C., “Personal authentication using palmprint features,” Pattern
Recognition 36(2), 371 – 381 (2003).
 Lowe, D., “Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision,
60(2), 91 – 110 (2004).
 Christmas, W, J., Kittler, J., and Petrou, M., “Structural matching in computer vision using probabilistic
relaxation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 17, 749 – 764 (1995).
 Kong, W., and Zhang, D., “Feature-level fusion for effective palmprint authentication,” In: Zhang, D., Jain, A.K.
(eds.) ICBA 2004 LNCS, 3072, 761 – 767 (2004).
 Kisku, D. R., Gupta, P., and Sing, J. K., "Feature level fusion of face and palmprint biometrics by isomorphic
graph-based improved K-medoids partitioning," 4th International Conference on Information Security and
Assurance Lecture Notes in Computer Science, 6059, 70 – 81 (2010).
 Han, Y.F., Sun, Z.N., and Tan, T.N., “Palmprint recognition based on directional features and graph matching,”
International Conference on Biometrics, LNCS, 4642, 1164 – 1173 (2007).
 Jain, A. K., and Jianjiang, F., “Latent palmprint matching,” IEEE Transactions on Pattern Analysis and Machine
Intelligence, 31(6), 1032 – 1047 (2009)
 Kostin, A., Kittler, J., and Christmas, W. J., “Object recognition by symmetrised graph matching using relaxation
labelling with an inhibitory mechanism,” Pattern Recognition Letters, 26(3), 381 - 393 (2005)
 Gold, S., and Rangarajan, A., “A graduated assignment algorithm for graph matching,” IEEE Transactions on
Pattern Analysis and Machine Intelligence, 18(4), 377–388 (1996).
 Ullman, J. R., “An algorithm for subgraph isomorphism,” Journal of ACM, 23(1), 31 – 42 (1976).
 Jornsten, K., and Nasberg, M., “A new Lagrangian relaxation approach to the generalized assignment problem,”
European Journal of Operational Research, 27, 313 – 323 (1986).
 Guignard, M., and Kim, S., “Lagrangean decomposition: A model yielding stronger Lagrangian bounds,”
Mathematical Programming, 39, 215 – 228 (1987).