A Writer-Dependent Approach for Offline Signature Verification Using Feature Learning and One-Class Support Vector Machine

In many administrative and financial entities, signatures are frequently employed to confirm a person’s identification as well as the authenticity of various documents. Being a biometric characteristic, a signature is a fairly secure form of identification. Hence, efforts are being made to continually raise the bar for the security signatures, particularly by creating effective procedures for verifying signatures. Both traditional and automatic techniques have been used to verify signatures, whiles the former has recorded moderately accurate results, the latter in recent studies have demonstrated great success in separating authentic signatures from forgeries. The purpose of this work is to explore the implementation and deployment of offline writer-dependent signature verification systems, by proposing a model that can greatly minimize computation while simultaneously increasing accuracy, to achieve this, we propose combining the VGG-16 architecture as a feature extractor with a One-Class Support Vector Machine (OC-SVM) as a writer-dependent classifier.