Real-Time Bit-Level Encryption of Full High-Definition Video Without Diffusion

1. Problem Statement

Current video encryption algorithms predominantly rely on the Shannon Confusion-Diffusion Architecture (SCDA). While effective for security, this architecture faces critical bottlenecks in real-time high-definition (HD) video applications:

• Sequential Dependency: Diffusion operations require establishing inter-pixel dependencies sequentially to ensure plaintext sensitivity. This inherent sequentiality severely limits algorithmic parallelism, preventing optimal utilization of hardware resources (specifically GPUs).
• Computational Overhead: To meet strict plaintext sensitivity metrics (resistance to differential attacks), multiple rounds of confusion and diffusion are often required. This incurs excessive computational overhead, resulting in encryption times that exceed the thresholds for real-time processing (e.g., >33ms for 30 FPS video).
• Latency: Existing methods, even those utilizing parallel computing, struggle to encrypt Full High-Definition (1920×1080) video at 30 FPS without introducing significant latency or frame drops.

2. Methodology

The authors propose a Real-Time Video Encryption Protocol based on Heterogeneous Parallel Computing that eliminates the diffusion phase entirely. The core innovation is leveraging the extreme input sensitivity of the SHA-256 hash function to achieve plaintext sensitivity without sequential diffusion.

Key Technical Components:

• Heterogeneous Parallel Architecture:
  • CPU Side: A main thread ($T_m$) oversees the process, while multiple worker threads ($T_w$) concurrently compute SHA-256 hashes of sub-frames and generate encryption data (shift distances and random bytes) using Lorenz Hyper Chaotic Systems (LHCS).
  • GPU Side: A dedicated GPU thread is assigned to every single pixel (across Red, Green, and Blue channels) to perform bit-level operations simultaneously.
• Diffusion-Free Sensitivity Mechanism:
  • Instead of using diffusion to propagate pixel changes, the protocol uses the SHA-256 hash of the original frame (or sub-frames) to reconstruct the initial conditions of the chaotic systems.
  • Because SHA-256 is highly sensitive to input changes, modifying even a single pixel in the original frame results in a completely different hash. This generates entirely different initial conditions for the chaotic systems, leading to distinct encryption keys and outcomes for the entire frame.
  • This approach achieves plaintext sensitivity with only one round of confusion and XOR operations.
• Encryption Workflow:
  1. Initial Condition Reconstruction: The SHA-256 hash of the current frame is used to reconstruct the initial conditions ($x_0, y_0, z_0, w_0$) for the Lorenz Hyper Chaotic System.
  2. Data Generation: CPU threads generate shift distances (for circular bit-shifting) and random bytes (for XOR operations) concurrently.
  3. Bit-Level Encryption (GPU):
     • Decomposition: Pixel values are decomposed into 8 individual bits.
     • Confusion: Bidirectional circular bit-shifts (horizontal and vertical) are applied to the bit matrix.
     • XOR: Bitwise XOR operations are performed between the shifted bits and the generated random bytes.
     • Reconstruction: Encrypted bits are reassembled into pixel values.

3. Key Contributions

1. Elimination of Diffusion: The paper successfully removes the diffusion step, which is the primary bottleneck for parallelism in video encryption. This allows for a 1:1 mapping of GPU threads to pixels, maximizing hardware utilization.
2. SHA-256 Driven Sensitivity: By utilizing the avalanche effect of SHA-256, the protocol achieves high plaintext sensitivity (resistance to differential attacks) with a single round of operations, drastically reducing computational load.
3. Dynamic Key Space: The key space is not fixed; it dynamically expands based on the number of frames encrypted ($S_k = (5 \times 64) + n_f \times 256$), providing robust resistance against brute-force attacks.
4. Mitigation of Dynamical Degradation: The periodic reconstruction of chaotic system initial conditions using frame hashes prevents the system from falling into short cycles or degraded states common in finite-precision digital implementations.
5. Robustness to Data Loss: The absence of diffusion means that errors (cropping or noise) in the encrypted frame do not propagate to other pixels during decryption, allowing for high-quality recovery even with damaged data.

4. Experimental Results

The protocol was tested on various platforms, including a high-end server (Intel Xeon Gold 6226R + NVIDIA RTX 3090), a desktop PC, and an embedded system (NVIDIA Jetson Xavier NX).

• Performance:
  • Achieved delay-free encryption for Full HD (1920×1080) video at 30 FPS on the server, with an average encryption time of 25.84 ms.
  • Achieved real-time encryption for 640×480 video at 24 FPS on the embedded Jetson Xavier NX (28.76 ms).
  • Outperformed recent state-of-the-art algorithms (Refs [5]–[11]) in both speed and resolution capabilities.
• Statistical Properties:
  • Uniformity: Encrypted frames showed uniform histograms with $\chi^2$ values well below the critical threshold (293.25).
  • Correlation: Adjacent pixel correlation coefficients were reduced to near zero (approx. 0.008) compared to ~0.99 in original frames.
  • Entropy: Information entropy approached the theoretical maximum of 8 (avg ~7.999), and local Shannon entropy passed the rigorous statistical test.
• Security Analysis:
  • Differential Attacks: Achieved NPCR > 99.6% and UACI within the critical interval (33.37%–33.55%), confirming high sensitivity to single-pixel changes.
  • Brute-Force: The dynamic key space exceeds $2^{100}$ after just one second of encryption.
  • Noise/Cropping: The system successfully decrypted frames with up to 37.5% data loss (cropping) and significant salt-and-pepper or Gaussian noise, restoring the original image with high visual fidelity.

5. Significance

This work represents a paradigm shift in video encryption by proving that diffusion is not strictly necessary for achieving high security and plaintext sensitivity if the key generation mechanism is sufficiently sensitive (via SHA-256).

• Real-World Applicability: It enables the first known implementation of bit-level, real-time encryption for Full HD video on standard hardware, making secure video transmission feasible for military, traffic monitoring, and social networking applications without latency.
• Scalability: The heterogeneous parallel approach scales effectively from embedded systems to high-performance servers, ensuring that secure video can be deployed in resource-constrained environments (like IoT cameras) as well as data centers.
• Efficiency: By reducing the encryption process to a single round of confusion and XOR, the protocol significantly lowers the computational barrier for secure video, potentially democratizing high-security video transmission.