Offline Signature Verification Based On Direction Features

With the high speed development of information technology, personal information security has been paid more and more efforts by researchers and companies. As an important way of biometric identification, offline signature verification can be effectively used in the personal identification. Some important steps of signature verification such as preprocessing, feature extraction, verification are introduced in this paper.Mathematical morphology methods are used to denoise binary signature images. The contour of each signature image not only contains the geometric structure features, but also describes the signature strokes. Features we extracted are based on the contour of signature images.In this paper, we employ the direction feature and the statistical feature. The direction feature mainly records the direction and the location of the special strokes. This feature is very sensitive for tiny changes of images. On the other hand, the statistical feature focuses on statistical features of signature images. It is not so sensitive for some small changes. Therefore, we fuse these two kinds of features together. Then we can get the high dimensions features which can better describe the signature images.After feature extraction, we use principal component analysis to reduce redundant features. After the dimension reduction processing, we can get lower dimensions features which can keep the principal component information. With lower dimensions, the training and classification speed can also been highly improved.In this paper, support vector machine is adopted as our classifier. Experiments on GPDS and MCYT databases demonstrate that the combination of the direction feature and the statistical direction feature is effective to improve signature verification’s performance.