Abstract of “Automated handwritten signature verification” final year project
Nowadays “Automated handwritten signature verification” is an active area of research these days due to its increased use by distinguished institutes, intelligence agencies, banks, etc. A signature is a very important attribute that is used to authenticate human identity. “Automated handwritten signature verification” is a challenging task.
Even if two signatures are made by the same person their properties may vary. Therefore, detecting forgeries, especially those done by skilled forgers is a very difficult task. In this project, we have researched pre-existing systems for “Automated handwritten signature verification” to find out which methods produce better results. Furthermore, an android based application solution is proposed that verifies the authenticity of signatures.
The final year project is developed such that it provides an easy-to-use interface and accurate results instantly. The distinction between real signatures and forged signatures is achieved through digital image processing techniques and machine learning algorithms. The dataset used for training purposes is first preprocessed to bring the data into a more usable form. This dataset is then divided into real and forged signatures.
The machine algorithms used for training the model are multilayer perceptron, convolution neural networks, and support vector machine.
The outcome produced by each model is evaluated and it is clear that poor performance is achieved via MLP. SVM produced better results with an accuracy of about 92%. However, the best results are achieved by using CNN with an accuracy of about 98%.