Stamp Verification

Abstract

Stamps have become one of the most important security features in big companies where huge amounts of documents need processing everyday. The stamp attached to a document is used to determine the authenticity of that document so that it is necessary to identify whether a stamp is forged or genuine. However, because of the widespread use of high-quality color copiers, it is now possible to produce forged papers with forged stamp images that are easily mistaken for genuine stamps. This created a big risk for many companies in data security and documents authentication. Therefore, an automatic system for stamp verification needs to be developed to deal with this problem. This thesis presents a practical approach for stamp verification, based on the 3 stages process similar to some previous work: Stamp segmentation, classification stamp or non-stamp and stamp authenticity verification. In each stage, this thesis tries and tests new algorithms/methods to give a new way of solving the problem in each stage. Firstly, in our approach, an unsupervised learning machine method is implemented to detect all the objects in the input image, so all the regions including stamps and text are extracted. Next, two separate models of Support Vector Machine classification are constructed. The first one is to distinguish between stamps and other objects in a document. The second model will determine the object which was classified as stamps in the first model whether it is genuine or not. The results show that this approach can perform the stamp verification tasks effectively.