We want no one to have to have to fall victim to identity theft. In 2014, 16.4 million persons annually in the US suffer from signature forgery alone. The signature forgery detection system, implemented, will be able to greatly reduce the number of successful forgeries–as shown in the results section of this report–granting individuals, businesses, and banks a greater piece of mind during transactions.
Begin with the MNIST dataset to prove capability to create a successful classification network using the Matlab’s Neural Network Toolbox
Use scanned signatures to implement image preprocessing steps to best focus the network on the signature resulting in a high SNR.
Obtain a few hundred signatures from ten unique people to train and test the network
Optimize the signature network
Utilize a physical device capable of capturing a new signature and running it through the network to verify if the signature is forged or not