RollingEvidence: Securing Video Authenticity with Rolling Shutter Effect

Addressing Video Manipulation Challenges

• ⚠️ AI-driven manipulations like deepfakes threaten the integrity and authenticity of video evidence vital for legal, security, and justice applications.
• 💡 RollingEvidence is a novel system designed to create and verify authentic videos by embedding invisible probes during recording.

How RollingEvidence Works

• 🎥 The system utilizes the rolling shutter effect in CMOS cameras, which causes row-by-row scanning, to embed real-time probes.
• ⚡ A modulated LED produces frequencies unseen by the human eye, which, combined with the camera's exposure, forms unique stripe patterns in frames.
• 🔑 It employs autoregressive encoding, linking frames to prior ones and incorporating device-specific cryptographic keys for robust tamper detection.
• 📊 Frequency Shift Keying (FSK) with 4,096 probe permutations is used, surrounded by splitter frequencies to enhance detection accuracy and enable compact, high-dimensional probe definitions.

Verification and Tamper Detection

• 🧠 During verification, custom deep neural networks extract these stripe patterns and decode the embedded probes.
• ✅ Tampered frames are then detected using exponential-min implication by comparing decoded probes against random sequences.
• 🖼️ The system also generates strip-free videos for clear viewing, enabled by rapid LED switching that avoids persistent frame occlusion.

Security and Performance Validation

• 🔒 Theoretical analysis confirms the system's reliability, with attack vectors offering 128-bit to 256-bit security against sophisticated attackers.
• 📱 A prototype using a smartphone and Arduino Uno demonstrated high efficiency in detecting manipulations like insertion, removal, alteration, face swapping, and lip syncing.
• 📈 Experiments showed robust performance with 0% false acceptance rate and less than 0.5% false rejection rate for tempered videos, achieving 99.5% accuracy.

Key Contributions

• 🛠️ Development of a system for embedding tamper-resistant probes using the rolling shutter effect.
• 🧩 Introduction of autoregressive encoding for secure, compact probes tied to device keys.
• 🎯 Creation of a multitask deep network for efficient strip extraction, probe decoding, and tamper detection.
• ✅ Validation of the prototype through comprehensive experiments showcasing its efficiency and security in real-world scenarios.