Autenticacão de Assinaturas Online: Estudo dos Parâmetros do Dynamic Time Warping e da Representação da Assinatura

Abstract

Pattern recognition is an important aspect within the computational intelligence area, which helps solving problems that use computing tools. Among these problems we can cite face recognition, fingerprint identication and signature authentication. The relevance of automatic signature authentication is related to the recognition of an individual and his/her role in a complex system and it is often related to financial matters. This work presents a study of the Dynamic Time Warping parameters, which is an algorithm used to align two signatures and measure the similarity between them. In a first stage a set of experiments varied the main parameters of the algorithm in a broad range of values and the resulting averages of classification errors were evaluated. Based on these first evaluations the necessity to calculate dynamically one of these parameters, the gap cost,it was identified in order to adjust it for practical application. A proposal to calculate thisparameter is also presented and evaluated. It is also proposed and evaluated an alternative way to represent the signature attributes, considering the curvature at each point acquired in the acquisition process, using the normal vectors as a form of representation. The evaluations performed in the diverse stages of the study considered the Equal Error Rate (EER) as quality measure and the proposed techniques were compared to well-established ones, obtaining an average EER of 3.47 %.