Thermal cameras can be utilized inconspicuously to expose heat traces left on input interfaces, posing a rising threat of a new front for side channel attacks. This research project aims to significantly contribute to and build on previous studies on thermal attacks by investigating deep learning models that can improve the accuracy of thermal attacks and testing them in real-world scenarios in an attempt to understand the impact of thermal attacks on user privacy and security. As part of the evaluation of our deep learning model, we captured and annotated 1,500 thermal images to create the first dataset of thermal images that capture the heat traces following an interaction (i.e. password entries).

Funded by Engineering and Physical Sciences Research Council (EP/V008870/1) - Royal Society of Edinburgh (65040) - Taif University (TIU500)