Feature fusion using a modified genetic algorithm for face and signature recognition system
Combination of multi biometrics at feature level fusion is able to give more accurate classification result. This thesis focuses on the development of feature level fusion of bimodal biometrics system for face and dynamic signature recognition system The modalities of biometric are used due to the a...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2015
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/16629/16/Feature%20fusion%20using%20a%20modified%20genetic%20algorithm%20for%20face%20and%20signature%20recognition%20system.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Combination of multi biometrics at feature level fusion is able to give more accurate classification result. This thesis focuses on the development of feature level fusion of bimodal biometrics system for face and dynamic signature recognition system The modalities of biometric are used due to the ability to avoid spoof attack since it is difficult for impostor to imitate two different characteristics (behaviour and physical) at the same time. Most existing systems are dealing with feature fusion of the same domain such as image based of fingerprint and face. Thus, there is no issue of incompatible features to be fused compared to the proposed development. Balance of the combined features has not been assessed whereas it is essential to ensure one of the biometrics does not dominate accuracy performance. To overcome the issue of incompatible features to be combined, Wrapper Genetic Algorithm (GA) was implemented as the feature selection algorithm due to its ability to evaluate the features irrespective of which domain by masking the features with bit number. A
modified fitness function in Wrapper GA was introduced by adding a function to maintain the balanced of the selected features. Penalty's value was imposed to the function when there is imbalance occurs in the selected features. Therefore, the accuracy performance of this system based on the fitness function that will rely on
the percentage of correctly recognized samples and the balanced of selected features. Several approaches and benchmark data were used to validate the effectiveness of the proposed method compared to the unimodal system and normal feature selection method. Results show that the proposed method yield optimal recognition with the highest accuracy of 97.50%. In addition, the importance of both biometrics remains, while maintaining the balance of the selected features. |
|---|