Performance Analysis of Classifiers on Offline Handwritten Signatures

Signatures is one of the popular biometric that depicts or reveals the attribute of humans which are extensively being used to authorize and authenticate day to day needs. Signatures play a very crucial role since they are the most critical piece of information that are used by humans for financial and commercial applications and legal transactions. Even today we have been practicing the traditional culture of signing documents like cheques and legal papers. Therefore, it is becomes necessity to recognize these signatures accurately with minimal effort in an accurate manner which becomes tedious by using the traditional manual verification technique. Thus there is a requirement to develop an automated signature recognition system. Verification systems can be designed either as an offline or online system based on the application. This work here focuses on offline authentication system that accepts as image a scanned signature. Hence comes the need to perform pre-processing which on an offline system is too complex to implement since the stable dynamic characteristics would be missed out once a scanned image is used for analysis. Also since signatures are truly unconventional and mainly styled representations need arises to design a reliable signature authentication system that can correctly classify signatures to their appropriate classes namely genuine and forged. Finally the performance analysis has been carried out on the local and global datasets and the accuracy has been tabulated using two different classifiers. It is also observed in the work that intra personal variations are recorded and accuracy of 66% is obtained for linear classifier and 74% for logistic classifier.