ONLINE SIGNATURE VERIFICATION USING SUPPORT VECTOR MACHINE (SVM)

Abstract

Selection of features of subject to be identified is very important step before applying any classification technique as different features have different type of significance associated with them. In online signature verification scheme we have to take care of selecting those features that can discriminate forgery and genuine signatures by giving a clear classification boundary. Extracting some other features from existing features by means of some technique can improve importance of that feature. We have combined three approaches together to give an optimal set of features for signature classification. Mean and variance analysis is done to identify Global features having capability to discriminate genuine and forgery one. Some Global features of signature were selected by PCA helps in identifying genuine signature genuine. Converting local features into more reliable FAR reducing feature is done with DTW and extended regression technique. We have designed RBF neural network and used it for classification. Results from all variations in features and classifiers are observed and discussed. A combined feature set obtained from three methods is passed to SVM classifier and results were improved rather than selecting features from individual techniques. We have compared our results and some other related work that reported their results on SVC2004 and it is found that accuracy of our algorithm is 95.375% .