Intelligent biometric pattern password authentication systems for touchscreens

• We introduced several methods of pattern password authentication for touchscreens.
• We designed a touchscreen user interface and we collected touch durations.
• As the classifier algorithms we used ANN, ANFIS and RGB Histogram methods.
• 80 real attempts and 80 fraud attempts from 10 users are operated.
• We achieved EER of 8.75% for ANN, 2.5% for ANFIS, 7.5% for RGB Histogram.

Given the recent developments in alternative authentication interfaces for smartphones, tablets and touchscreen laptops, one of the mostly selected method is the pattern passwords. Basically, the users that prefer this method, draw a pattern between the nodes to open the lock in lieu of entering an alphanumeric password. Although drawing a pattern seems easier than typing a password, it has a major security drawback since it can be very easy to be stolen. Therefore, this paper proposes some novel theoretical ideas with artificial intelligence methods, to improve security of pattern password authentication, using touching durations as biometric traits. What we put forward is the utilization of three different neural network based algorithms to verify logins with one novel histogram-based technique in a hidden interface for enrollment, training and verification.Inspired by the keystroke recognition models, the touch time and durations are extracted to create a ghost password. Moreover, the nodes are colored depending on the touch duration in the hidden interface and subsequently the colored images are exported. As a result of training session, the system discriminates real attempts from frauds using artificial neural networks (ANN), adaptive neuro-fuzzy inference systems (ANFIS) and Red–Green–Blue (RGB) Histogram methods in verification phase. The results are greatly encouraging that we reached 0% of false accept rate (FAR) for 80 fraud attacks with 16.5% false reject rate (FRR) of unsuccessful authentication for the 80 real attempts when started with interval checking algorithm. Moreover, to reduce this FRR, we utilized neural network based systems and consequently with ANN, we achieved 8.75% equal error rate (EER), with ANFIS, 2.5% EER for 85% proximity and finally with RGB Histogram method, we attained 7.5% EER.