![Dakshina Ranjan Kisku, Phalguni Gupta, Jamuna Kanta Sing
Dept. of CSE, Asansol Engineering College, Dept. of CSE, IIT Kanpur, Dept. of CSE, Jadavpur University, India
**Contact: drkisku@ieee.org
Abstract Isomorphic Graph Representations:
This poster presents a feature level fusion of face and
palmprint biometrics. It uses the improved K-medoids
clustering algorithm and isomorphic graph. The performance
of the system has been verified by two distance metrics
namely, K-NN and normalized correlation metrics. It uses two
multibiometrics databases of face and palmprint images for
testing. The experimental results reveal that the feature level
fusion with the improved K-medoids partitioning algorithm
Exhibits robust performance and increases its performance
with utmost level of accuracy.
Steps:
• Detection and localization of face and palm image
• Extraction of SIFT feature points from face and
palmprint images
• Partitioning the SIFT points
• Establishing correspondence between feature points Fusion of Keypoints:
• Isomorphic graph representations
• Fusion of matching keypoints
• Matching
• K-Nearest Neighbor
• Correlation distance
SIFT Points Extraction:
Experimental Results:
SIFT Features Extraction from Face and
Palmprint Images
.
SIFT Points Clustering using Improve K-Medoids
Algorithm:
Step 1: Select randomly k number of points from the SIFT points
set as the medoids.
Step 2: Assign each SIFT feature point to the closest medoid
which can be defined by a distance metric (i.e.,
Minkowski distance over the Euclidean space)
Step 3: for each medoid i, i = 1, 2…k
for each non-medoid SIFT point j
swap i and j and
compute the total cost of the configuration
Step 4: Select the configuration with the lowest cost
Step 5: Repeat Step 2 to Step 5 until there is no change in the
medoid.
Improved version of PAM clustering using
Silhouette approximations:
( y (i ) + y (i + 1)) / 2 − ( x (i ) + x (i + 1)) / 2
S (i ) = August 19, 2010
max[(( x (i ) + x (i + 1)), ( y (i ) + y (i + 1))]](https://image.slidesharecdn.com/featurelevelfusionoffaceandpalmprintbiometrics-100817103352-phpapp02/85/S-SSPR-2010-Workshop-1-320.jpg)
S+SSPR 2010 Workshop
- 1. Dakshina Ranjan Kisku, Phalguni Gupta, Jamuna Kanta Sing Dept. of CSE, Asansol Engineering College, Dept. of CSE, IIT Kanpur, Dept. of CSE, Jadavpur University, India **Contact: drkisku@ieee.org Abstract Isomorphic Graph Representations: This poster presents a feature level fusion of face and palmprint biometrics. It uses the improved K-medoids clustering algorithm and isomorphic graph. The performance of the system has been verified by two distance metrics namely, K-NN and normalized correlation metrics. It uses two multibiometrics databases of face and palmprint images for testing. The experimental results reveal that the feature level fusion with the improved K-medoids partitioning algorithm Exhibits robust performance and increases its performance with utmost level of accuracy. Steps: • Detection and localization of face and palm image • Extraction of SIFT feature points from face and palmprint images • Partitioning the SIFT points • Establishing correspondence between feature points Fusion of Keypoints: • Isomorphic graph representations • Fusion of matching keypoints • Matching • K-Nearest Neighbor • Correlation distance SIFT Points Extraction: Experimental Results: SIFT Features Extraction from Face and Palmprint Images . SIFT Points Clustering using Improve K-Medoids Algorithm: Step 1: Select randomly k number of points from the SIFT points set as the medoids. Step 2: Assign each SIFT feature point to the closest medoid which can be defined by a distance metric (i.e., Minkowski distance over the Euclidean space) Step 3: for each medoid i, i = 1, 2…k for each non-medoid SIFT point j swap i and j and compute the total cost of the configuration Step 4: Select the configuration with the lowest cost Step 5: Repeat Step 2 to Step 5 until there is no change in the medoid. Improved version of PAM clustering using Silhouette approximations: ( y (i ) + y (i + 1)) / 2 − ( x (i ) + x (i + 1)) / 2 S (i ) = August 19, 2010 max[(( x (i ) + x (i + 1)), ( y (i ) + y (i + 1))]