A Genetic Algorithm Based One Class Support Vector Machine Model for Arabic Skilled Forgery Signature Verification.

Area of Science:

• Biometrics and Pattern Recognition
• Computer Vision
• Machine Learning

Background:

• Signature verification systems are crucial for security in forensic and commercial transactions.
• Current systems struggle with skilled forgery detection and require extensive training data.
• Scanned signatures often contain noise, complex backgrounds, and image degradation, complicating feature extraction.

Purpose of the Study:

• To develop an improved signature verification system that overcomes limitations of existing methods.
• To enhance the accuracy of skilled forgery detection.
• To address the challenge of limited training data in signature verification.

Main Methods:

• A four-step approach: preprocessing, multi-feature fusion, discriminant feature selection via a genetic algorithm (OCSVM-GA), and one-class learning.
• Utilized three signature databases: SID-Arabic, CEDAR, and UTSIG.
• Implemented a one-class learning strategy to handle imbalanced signature data.

Main Results:

• The proposed method significantly outperforms existing systems in signature verification accuracy.
• Demonstrated improvements in reducing the false acceptance rate (FAR), false rejection rate (FRR), and equal error rate (EER).
• Effectively handled noisy and degraded signature images.

Conclusions:

• The novel signature verification system offers superior performance, particularly in skilled forgery detection.
• The OCSVM-GA feature selection and one-class learning strategy effectively address data imbalance and improve robustness.
• This approach provides a more reliable solution for practical signature verification applications.