Personalized offline signature verification using multiple HMM-classifiers and SOM-fuzzy decision fusion
This paper presents a user-optimized multiple classifiers approach for an offline signature verification system. Local features are extracted from a sliding window that slides across the signature images. Multiple HMM-based classifiers are used for the soft decisions, where each classifier is train...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Scottish Group
2013
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/28817/1/Personalized%20offline%20signature%20verification%20using%20multiple%20HMM.pdf |
| Summary: | This paper presents a user-optimized multiple classifiers approach for an offline signature verification system. Local features are extracted from a sliding window that
slides across the signature images. Multiple HMM-based classifiers are used for the soft decisions, where each classifier is trained on a particular feature. In this
work, we select the two best features to represent each user via ANOVA statistical analysis. A fuzzy decision fusion system that is tuned using SOM based clustering technique is used to combine the soft decisions from the selected HMM classifiers in producing the final verification output. The system has been tested on SIGMA signature database which is a collection of over 6000 genuine and 2000 forged signatures. Results show that our
personalized multiple classifiers approach out performs common single classifier systems. |
|---|