Multiple Classifier System for Writer Independent Offline Handwritten Signature Verification using Elliptical Curve Paths for Feature Extraction

摘要

Various offline handwritten signature verification systems using writer independent approach are proposed by the researchers in last few years using numerous perspectives, like feature extraction techniques, feature selection techniques, classifiers used to develop the system etc. Despite the progressions in this framework, building classifier that can isolate the genuine and skilled forgery signatures is still a tough task. In this work, multiple classifier system is proposed to develop the writer independent offline handwritten signature verification system. To train the classifiers of multiple classifier system, feature vectors of the training set are partitioned into subsets and classifiers are trained using these subsets to preserve the diversity. The pixels lying on the elliptical curve paths are used to extract the features from genuine and forgery signature images. Two scenarios are proposed for the performance analysis. In the first scenario, the classifiers are trained using genuine signature and random forgery signature samples whereas genuine and all types of forgeries specifically random, unskilled and skilled forgery signatures are utilized for the training process of the classifiers in the second scenario. Signature database of 150 writers is used to perform the experiments. False rejection rate 8.33 and false acceptance rate 0.00, 0.00 and 1.67 for random, unskilled and skilled forgeries, respectively are reported as the best result of the experiments.