AUTOMATIC OFFLINE SIGNATURE VERIFICATION SYSTEM

Abstract

Automated offline verification system that verifies genuine signature and detects signature forgeries of any types is presented. The system used various algorithms to pre-processed signature images before presented for feature extraction. Scale Invariant Feature Transform was used as a feature extraction technique. Samples of three genuine signatures of writer were taken and their Scale Invariant Feature Transforms were extracted using MATLAB function. Euclidean distance was used to calculate variability within the same writer. This variability is computed as intra-class Euclidean distances. The feature vector Euclidean distances, the image distances and intra-class thresholds are computed and stored as templates for a known writer. For any test signature scale Invariant Feature transforms is extracted and inter-class Euclidean distances is calculated, that is, the Euclidean distances between feature vectors of test signature and those of stored template. The intra-class threshold stored in the template is compared with the inter-class threshold for the test signature to be considered as authentic or forgery. The system was implemented on a database of 140 signatures consisting of training set and test set. The system is not only able to verify genuine signature but also detects all types of forgeries (Random, unskilled and skilled).