A biometric identification system
11.03.2003.
The multiplatform system for automatic person identification, based on the fusion of palm, fingers and hand geometry features.
Key features
- a fusion of palm and hand geometry features
- a single sensor is used for extracting palm and hand geometry features (fusion at the feature extraction level).
- no pegs, the user has to put his hand on the input device with fingers spread naturally.
- translation and rotation of the hand is allowed.
- spatial resolution of the images is 180 dpi / 256 grey levels.
- a fusion at the confidence level (accept/reject decision).
- at the fusion level the neural network is used
- system was written in C++ and tested on Linux and WIN32 platforms. GUI for demo program was written using FOX toolkit.
Experimental results
- The biometric identification system was tested using a database of 112 persons who were not used in the training phase.
- At least five images of each person's hand were captured, thus more than 560 images were made available. In the enrolment phase we used three images, and so for testing the users there were at least two available images. For the impostors at least five images were available for testing.
- There was more then 900000 matching between samples during the experiment.
- False Reject Rate (FRR) and False Accept Rate (FAR):
Learn database
(54 persons)Test database
(112 persons)Low
thresholdFRR [%] 0.00 0.38 FAR [%] 0.00 0.00 Normal
thresholdFRR [%] 0.00 0.57 FAR [%] 0.00 0.00 High
thresholdFRR [%] 0.79 1.90 FAR [%] 0.00 0.00 - Execution time:
Feature
extractionIdentification Verification Linux 130 ms 60 ms < 1 ms MS Windows 2000 200 ms < 1 ms < 1 ms
Hand samples
Hand samples with extracted finger features and palm area.
[ Full size images will be open in the new window. ]
Screenshots
[ Full size images will be open in the new window. ]