MLP-Based Attribute Selection Method for Handwritten Signatures Authentication

Abstract

Finding the most unique traits that have strong discrimination capacities to be used for confirmation, in particular with reference to the substantial variation that’s intrinsic in real signatures, is among the main difficulties in developing an algorithm for electronic signature validation. Handwritten signs offer the potential for expertly made frauds that closely resemble genuine equivalents. During this work, we proposed a methodical approach for authenticating online signs via an MLP that relies on a predetermined set of PCA (principal component analysis) features. This suggested method demonstrates an attribute selection methodology using data obtained from PCA calculations that is often disregarded but may be important in achieving a lower error rate. Utilizing a 5000-sign sample from the SIGMA database, the study produced false rates of acceptance (FAR) and false rates of rejection of 17.4% and 16.4%, respectively.