E-Meet (or Emit?) My Keystrokes: How Benign Screen-sharing Meetings Could Leak Typing Behaviors
summary
Computer vision algorithms recover keystroke timing and error patterns from ordinary screen‑sharing recordings, not camera footage, by tracking the blinking text cursor and pixel changes; extraction uses frame differencing at 30 fps ≈33 ms per frame, probabilistic character-box scoring, and behavioral matching to build a typing profile. Results show per‑character identification peaks around 70% per‑character detection (degrades with faster typists), enable imitation without malware (no keylogger needed), and include a working extractor and evaluation on real meeting recordings.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 8:35Screen-sharing risks
- 11:59Typing activity detection
- 16:58Character extraction challenges
- 28:59Accuracy limitations
- 39:18Future attack possibilities
related talks
24mSecurity at your fingertips: Theory → Practice!
30mCommSec D1 - Keynterceptor: Press Any Key to Continue
23mExploring Intentional Behaviour Modifications for Password Typing on Mobile Touchscreen
24mMy Google Glass Sees Your Passwords!
37mWindows Keylogger Detection: Targeting Past & Present Keylogging Techniques
40mVideo-Based Cryptanalysis: Recovering Cryptographic Keys from Device Using Video of Power LED
37mHi, My Name Is Keyboard
25mThermanator and the Thermal Residue Attack