No abstract

On mobile devices, the most important input interface is touchscreen, which can transmit a large amount of sensitive information. Many researchers have proven that sensors can be used as side channels to leak touchscreen interactive information. The research of information leakage in the restricted area has been relatively mature, but in the unrestricted area, still there are two issues to be solved urgently: chirography difference and posture variation. We learn from the way spiders perceive prey through the subtle vibrations of their webs; an unrestricted-area handwriting information speculation framework, called spider-inspired handwriting character capture (spider-inspired HCCapture), is designed. Spider-inspired HCCapture exploits the motion sensor as the side-channel and uses the neural network algorithm to train the recognition model. To alleviate the impact of different handwriting habits, we utilize the generality patterns of characters rather than the patterns of raw sensor signals. Furthermore, each character is disassembled into basic strokes, which are used as recognition features. We also proposed a user-independent posture-aware approach to detect the user’s handwriting posture to select a suitable one from some pretrained models for speculation. In addition, the Markov model is introduced into spider-inspired HCCapture, which is used as an enhancement feature when there is a correlation between adjacent characters. In conclusion, spider-inspired HCCapture completes the handwritten character speculation attack without obtaining the victim’s information in advance. The experimental results show that the accuracy of spider-inspired HCCapture reaches 96.1%.