Keystroke Inferencing Attack Revealed: What You Need to Know
Secretly filmed you playing with your phone from a distance of 12 meters, and successfully recovered the content you entered on your phone đź‘»
Researchers at the University of Chicago have developed a new attack method that uses a less than $60 telephoto lens mounted on a smartphone to photograph a victim typing from inside a building (behind a window) about 12 meters away, successfully recovering The content being entered.
This method requires no prior training, keyboard knowledge, local sensors or side channels, etc.!
Attack overview:
1. Nature of the attack: This is a video-based keystroke inference attack that can be performed in public places.
2. Use equipment: The attacker only needs an ordinary RGB camera to shoot the target’s typing fingers from the front.
3. Unique method: Unlike previous methods, this attack does not rely on side-channel data or other assumptions, only a frontal view of the target’s typing hand. It does not require pre-training, keyboard knowledge, training data for the target, local sensors or side channels.
4. Sample scene: In an indoor lounge scene, the attacker records the victim’s typing movements while watching the video. Long-distance outdoor scene, where the attacker uses a smartphone with a cheap telephoto lens to film the victim typing in the courtyard from about 12 meters away.
5. Diverse conditions: Attacks are evaluated under different conditions, including different environments (indoor/outdoor), attack distances, obstacles, and keyboard devices (visible/invisible keyboards, different sizes/layouts).
6. User study: The study involved 16 different users with different typing styles and abilities. Aggression showed high effectiveness in almost all scenarios and performed well among participants with significant behavioral differences.
Main technical principles:
- Video analysis
Hand tracking: First, use video analysis technology to track and analyze the finger movements of the target person.
Keystroke detection: Detect keystrokes by analyzing finger movements and position changes.
2. Data processing
Self-teaching system: Use a two-layer structure self-teaching system to process video data. This system consists of two main parts: keystroke detection and clustering: using the results of hand tracking to detect keystrokes and classify them.
Hidden Markov Model (HMM): Use HMM to identify specific keystrokes.
3. Infer the input content
Language model: Combined with the language model to analyze and infer the keystroke sequence to infer the input content.
3D-CNN model: Use a 3D convolutional neural network (CNN) model to further process the data and improve the accuracy of inference.
Detailed introduction: https://t.co/wautv7NDLM
paper:
More AI News
Artificial Intelligence Article
New AI Technology
