A CNN-based keylogger using acceleration spectrograms
Palavras-chave:
Deep learning, Signal processing, Convolutional neural networks, Keylogger, Keyboard attacksResumo
This work presents a model of supervised Convolutional Neural Network to identify keys pressed in a
computer keyboard based solely on their acceleration signal. In this problem, vibration propagated through a desk
from keystrokes were captured by an accelerometer. Individual keys have different acceleration signatures, and the
network was used to tell keys apart by looking at their particular acceleration spectrograms. The proposed network
showed an accuracy of up to 92% in the identification of individual keys.
