School: Desert Academy
Area of Science: Computational Science
Interim: The goal of this project is to create a program that can identify whether or not a signature is forged or not. The main application for this is to add another layer of security to things that require signatures like credit cards, and Atms. We are planning on doing this with artificial intelligence in the form of neural network. We have been using Tensor Flow, an open source google library written in python. So far we have spent most of our time developing a method of collecting data to train out network. We started out using a drawing tablet and a simple processing script that allowed the user to use the drawing tablet to write their signature. When the signature is completed the program output a text file with a 1 dimensional list of the pixels in the signature’s image. This worked great but a problem arises when we tried to get people to use it. Do the high learning curve of a drawing table people spent a lot of time trying to learn how to use the tablet. Also if they managed to use the tablet there signature was inconsistent and messy. Realizing this we decided to use the simplest thing we knew. Pen and paper. We printed out charts in which people signed their name multiple times. We scanned the sheets and began to split the image file into a different image file for each signature using adobe Photoshop. We still are getting more people to sign and are working on processing script to automate the splitting of the image files. We have also started on the actual neural network. We have made a proof of concept using the mnist fashion data set to see if comparing images reliability in possible. It works but is just a proof of concept and is at a much smaller scale. Over all we have made a lot of progress toward our goal. What we have left to do is complete collecting data and make a user friendly interface for the network.
Sponsoring Teacher: Christopher Zappe
Mail the entire Team