SketchyPass

Sketchy Pass is a new type of password for use on smartphones, which is an improvement on the "Pattern Unlock" feature that is currently in use. Instead of storing a password as a pattern on a specified grid, we are going to allow the user to draw a freeform sketch on their phone without a grid. This has the potential to improve both security and usability. Sketchy Pass is more secure because it allows for a broader password space, and it is more resistant to screen smudge exploitations because the freeform drawings produce more natural shapes that are harder to detect and reproduce. It is also more user-friendly than pattern unlock because object-based sketches are easier to remember than abstract patterns.

Method

The first step in our project will be to reduce the dimensionality of our images to decrease the complexity of our calculations. We will do this with principal component analysis and multidimensional scaling. We may also normalize the coordinates of our images to allow for translations of the same sketch. After the sketches have been pre-processed, we will develop an appropriate metric for image distance using unsupervised, parametric clustering. The task of determining whether the new sketch matches the password sketch can then be approached as a binary classification problem using MAP logistic regression.

Dataset

The training and test data for this project will be supplied by volunteers who will draw a password sketch 5-10 times in an Android app. Roughly 70% of the sketches will be used as the training set, while the remaining 30% will be held out to be used as the test set.

Milestone Goal

By the milestone (February 19th), we hope to have at least 100 volunteers provide 5-10 sketches each via the Android app. We also hope to have finished dimensionality reduction, implemented unsupervised, parametric clustering, and developed an initial image distance metric.