2D correlation for signature matching

Abstract

Handwritten signatures are widely used in security and financial transactions, making signature authentication systems important. Signature verification is challenging as nobody signs identically. Differences can include strength and number of strokes, shape and geometric transformations. An off-line matching system that is invariant to subtle dissimilarities and yet able to distinguish between signatures is needed.A 2D-Correlation Signature Matching System has been designed. This system utilizes regional correlation to extract unique templates from original signatures for comparison on target signatures. Geometric transformation functions were also included. These make the system less susceptible to noise, quality and minor differences in signatures. Its simplicity and adaptability reduced its learning time and complexity over present techniques. Although computational intensive, this is easily offset by faster computers and efficient image processing libraries.The system has been evaluated with 224 signature samples and the results show that an average error rate of 1.88% with 1.98% false rejection and 1.78% false acceptance rates have been achieved.