A Novel End-to-End Approach for Offline Signature Verification System

Recently, security application and demand is increasing rapidly. In this field of security, bio-medical security plays important role which include fingerprint verification, iris verification etc. However, signature verification is also considered as a promising technique which can help to improve the security system by verifying the appropriate user based on their signature behavior. These systems are widely adopted in banking system for cheque verification and transaction process. These type of security systems suffer from various security threats which can be caused by signature forgery. To address this, various researches have been presented for offline signature verification system but achieving promising accuracy performance is still considered as a challenging task in this research field. In this work, we have focused in the offline signature verification system and developed a complete end-to-end model for signature verification system by following various stages such as image denoising, feature extraction, database training and classification. First of all, image denoising technique is implemented using markov random field, next phase of work presents combined feature extraction process using SIFT and LBP feature extraction. Later, optimized support vector machine based database training and machine learning technique is presented. An extensive experimental study is presented on the GPDS dataset by dividing it into various test cases where separate stages of this work are implemented and performance is evaluated. Study shows that complete proposed approach achieves better performance when compared with state of art models.