A novel approach to signature verification and identification

In this paper we are exploring a novel approach to extracting the features from a hand-written offline signature. The experiments are carried out on a user created data base. Which includes 50 classes and each class consists of 15 random forgeries and 5 skilled forgeries; total count of samples in the data base is 1000. Here we are considering and extracting the geometrical distance metric features, end points alignment and pruned projection features, vector of angle and one existing SIFT features. To obtain the similarity matrix we apply the different technique called one-to-one string pattern matching. The simple measure of similarity R itself is serve as a feature. Late the patterns are classified using one supervised classifier like Knn-classifier the results are compared. In biometric experiments accuracy should be measured in terms of equal error rates. We are comparing the results of existing and proposed methods and justifying the results. For verification and identification error rates are 1.2% and 1.4% by novel approach and that of existing SIFT feature is 1.4% and 1.6%.