IDENTITY AUTHENTICATION BASED ON HANDWRITTEN SIGNATURE USING FUZZY CLASSIFIERS ENSEMBLE

Authentication applications based on the use of biometric methods have received great interest in recent years. Authentication is a two-class classification problem. Fuzzy classifiers and their ensembles made a breakthrough in many different pattern recognition tasks. However, using these systems with a traditional objective function based on overall accuracy does not work properly when the amount of training data is insufficient. This refers to the task of online signatures verification. This article proposes a new objective function based on the differences of the compatibility degree between the pattern and each fuzzy rule. At the beginning, handwritten signature signals are preprocessed, including eliminating gaps, eliminating tilt, normalizing position, and scaling. The paper investigates the effectiveness of a fuzzy classifier in the ensemble. To build an ensemble of classifiers, a well-known bagging method is used. Experiments on recognizing the signature using SVC2004 and MCYT-100 databases with the construction of a single fuzzy classifier and ensembles of three, five, seven and nine fuzzy classifiers were conducted. The experiments were carried out both with the use of a traditional objective function and with the use of a new objective function. Extensive evaluation of the fuzzy ensemble constructing method against the recent models shows that the present approach is efficient in recognition and verification of signatures.