Offline Signature Verification Based On Improved Scale Invariant Feature Transform

Signature for the author's authentication has hundreds of years of history.Signature is used in commercial contracts,bank loans and many other areas.It is very important to improve the security of the authentication system and avoid the economic loss.Compared with online signature,off-line signature does not have dynamic information such as velocity,pressure and stroke sequence.The artificial off-line signature verification is not only inefficient,but also easy to be interfered by subjective impression.The computer can effectively improve the efficiency and accuracy of verification.The skilled forged signatures and scale change of signature images have great influence on the verification result.To solve the problem,this thesis proposes a novel method based on the scale invariant feature transform for off-line signature verification.The main work of this thesis is as follows:(1)Due to the characteristic of SIFT feature,we only need to change signature images to gray and remove background noise in preprocessing stage.The signature image has simple background and the most image information is focused on the character.Too many octaves and layers in scale space may increase the computational complexity of the algorithm.Through the experimental analysis,the number of layers and the number of octaves are reduced.(2)The SIFT feature points are matched through Euclidean distance.Select the matches through the ratio of the distances,difference of the feature point orientation and RANSAC.The number and average distance of reserved matches are as the parameter of verification.The two parameters can achieve the preliminary verification of off-line signatures.(3)The number and average distance of reserved matches just can verify a part of skilled forged signatures.So we extract ODH feature vector which is composed of histogram statistics on orientation difference of the feature points.Manhattan distance is used as the similarity of ODH feature vector.In order to fully utilize the SIFT feature point,we use the bag of words model to extract BOW feature vector.Through ODH feature vector and BOW feature vector,we can more accurately verify the skilled forged signatures and improve the accuracy of verification.(4)In this thesis,we use a local offline signature data set and 4NSigComp2010,SigComp2011 two offline signature public data sets.The equal error rate is as the evaluation of the experimental results.The EER on the local data set is 6.2%.The EER on the 4NSigComp2010 data set is 20% and on the SigComp2011 dataset is 6.9%.The experimental results show that the method has obtained good results in the verification of offline signatures.