Invariant directional feature extraction and matching approach for robust off-line signature verification

Off-line handwritten signature verification systems are highly required in different domains like financial transactions, forensic sciences, and legal document authentication. The proposed directional features are extracted from off-line signatures to accurately describe the angle of the pixels connection in the signature, for the reason of differentiating accurately between genuine and skilled forged signatures. Directional feature requires the adjustment of the skewness of the signature to uniform axis in the preprocessing of the signature. The accuracy of signature rotation varies according to the degree of the intra-personal variations of the signer. This work proposes a feature extraction method that generates rotation invariant directional features. The angle of each connection between pixel-pair is considered with respect to the next pixel-pair connection rather than to a global signature axis. Then matches between two different signatures by counting the longest common subsequence in these signatures. The experimental work are applied on two benchmark data sets that include skilled and genuine signatures. The accuracy of the proposed approach is between 90%-94%, outperforms the traditional SIFT/SURF based approaches.