Signature recognition by using SIFT and SURF with SVM basic on RBF for voting online

The Signature recognition is known as the process to verify a writer by examining the signature upon samples has been studied and stored in the database .This process has two types: The offline and the online. This thesis deals with the offline technique and proposed a SIFT and a SURF algorithm which is used to detector and descriptor keypoint (features) for each signature image. This process, Bag-of-word features, is operated by making vector quantization technique, which is outlined the key points for each training image inside a unified dimensional histogram. Features of bag-of-word are put inside multiclass Support Vector Machine (SVM) classifier established upon the Radial Basis Function (RBF) for a training and testing. Open CV C++ is used as an image processing tool and tool for feature extraction. In this thesis, the performance of SIFT on SVM based RBF kernel is compared with SURF on SVM based RBF kernel .It is found that the use of SIFT with SVM-RBF kernel system, has an accuracy of 98.75% and SURF with SVM-RBF kernel has an accuracy of 97.5%.We used TCP/IP that uses the client/server model of communication in which a computer user requests and is provided a service by another computer in the network. The client server network communication system is implemented by using java programming language for application interface and use to detect the network feature and use java serializable interface method to store the image of the system and important data of the system. Key Words: Signature recognition, SIFT, SURF, SVM-RBF kernel, BOW (bag of words), Open CV C++, TCP/IP.