Syntax Element Encryption for H.265/HEVC Using Chaotic Map-Based Coefficient Scrambling Scheme

Technical Summary: Syntax Element Encryption for H.265/HEVC Using Chaotic Map-Based Coefficient Scrambling Scheme

Problem Statement
High-efficiency video coding (H.265/HEVC) is the dominant standard for video transmission, yet the sensitive nature of streaming data necessitates robust privacy protection. While Selective Encryption Algorithms (SEA) offer a balance between security and computational efficiency by encrypting only specific syntax elements, existing methods face significant limitations. Current coefficient scrambling techniques often impose heavy computational loads, result in excessive bit rate overhead due to encryption-induced statistical changes, and suffer from insufficient safety measures against statistical and differential attacks. Furthermore, many existing schemes struggle to maintain format compliance or fail to adequately distort video content without compromising decoding integrity.

Methodology
The paper proposes a novel encryption scheme that integrates syntax element encryption (SEC) with a chaotic map-based coefficient scrambling mechanism. The approach operates synchronously with H.265 slices to ensure resilience against packet loss.

1. Random Sequence Generation: The scheme utilizes AES-CTR to generate random sequences. To resist chosen-plaintext attacks (CPA), the initial key and initial vector are dynamically updated for each slice using the SHA-384 hash value of the slice header. This ensures that every slice possesses a unique random sequence.
2. Syntax Element Encryption: The method selectively encrypts five specific syntax elements to induce significant visual distortion while maintaining format compliance:
   • Motion Vector Difference (MVD): Encrypts the sign flags and the suffix of the absolute values.
   • Quantized Transform Coefficients (QTC): Encrypts the sign flags and the suffix of the absolute levels, subject to specific conditions to prevent Rice parameter violations.
   • Luma Intra Prediction Mode (Luma IPM): Encrypts the prediction mode using a modular addition operation. Crucially, the scheme includes logic to adjust Chroma IPM and coefficient scan modes based on the encrypted Luma IPM to maintain bitstream validity.
3. Chaotic Map-Based Coefficient Scrambling: To disrupt coefficient distribution effectively, the authors employ a Logistic-Sine Chaotic Map (LSCM). This low-dimensional discrete chaotic system generates pseudorandom values to determine the scrambling indices.
   • Scrambling Conditions: The scheme introduces specific conditions to exclude certain coefficient indices (e.g., coefficients with a value of 1 or specific leading coefficients) from scrambling. This design aims to preserve video edges and minimize the bit rate overhead typically caused by coefficient perturbation.
   • Process: The scrambling involves sorting pseudorandom values to generate indices, determining the number of shifts based on the random sequence, and applying right-shift perturbations to the coefficients.

Key Contributions

• Novel Scrambling Scheme: The paper introduces a coefficient scrambling method based on the Logistic-Sine Chaotic Map that leverages pseudorandomness and sensitivity to initial conditions to disrupt coefficient distributions more effectively than traditional exchange or shift methods.
• Optimized Bit Rate Overhead: By designing specific exclusion conditions for coefficient scrambling, the scheme significantly reduces the bit rate overhead compared to existing methods like the Robust Selective Video Encryption (RSVE) scheme.
• Enhanced Security Architecture: The integration of SHA-384 for key updates and AES-CTR for random sequence generation provides resistance against chosen-plaintext attacks. The slice-synchronized operation ensures that video decryption remains functional even if specific slices are lost during transmission.
• Comprehensive Evaluation: The study provides a theoretical proof of security against CPA and a detailed experimental evaluation across multiple performance metrics, including security, format compliance, and efficiency.

Experimental Results
The proposed scheme was evaluated using 12 video sequences with resolutions ranging from 352×288 to 2560×1600, comparing performance against state-of-the-art methods (including RSVE, Jovanovic et al., and Sallam et al.).

• Security and Distortion: The encrypted frames exhibit high visual distortion, making the original content unrecognizable. The scheme demonstrates strong resistance to substitution attacks and chosen-plaintext attacks, with Average Hamming Distance (HDavg) values close to 0.5. Notably, the proposed scheme achieved the lowest PSNR values among all compared methods, indicating the highest level of distortion and superior confidentiality for encrypted video content.
• Bit Rate Efficiency: A primary finding is the reduction in bit rate overhead. The proposed scheme achieved an average bit rate overhead improvement of 45.13% compared to the RSVE scheme, with specific video sequences showing improvements ranging from ~23% to ~67%.
• Execution Time: The average execution time overhead was reduced to 1.91%, significantly lower than the ~17.05% overhead observed in the RSVE scheme, indicating high computational efficiency.
• Quality Metrics: The scheme achieved low Peak Signal-to-Noise Ratio (PSNR) values for encrypted frames, confirming the highest distortion level among competitors, and high PSNR for decrypted frames, indicating accurate reconstruction. However, the Structural Similarity (SSIM) for encrypted frames was noted to be lower than some competing methods, and the Information Entropy of the encrypted frames was slightly lower than the RSVE scheme in some cases.
• Edge Preservation: Edge Difference Ratio (EDR) analysis showed values close to 1, indicating effective hiding of edge information.

Significance and Claims
The authors claim that this work presents a superior option for H.265/HEVC video security by effectively balancing the often conflicting requirements of high security, format compliance, and low overhead. The significance of the work lies in its ability to:

1. Minimize Overhead: It addresses the critical issue of bit rate escalation in selective encryption, offering a solution that is significantly more efficient than the benchmark RSVE scheme.
2. Ensure Robustness: The slice-synchronized design ensures reliability in real-world network conditions where packet loss may occur.
3. Provide Theoretical and Practical Validation: The paper offers formal proofs regarding resistance to chosen-plaintext attacks and chaotic system dynamics, backed by extensive empirical data.

The authors acknowledge limitations, noting that while the scheme excels in bit rate efficiency and encryption speed, it shows slightly lower performance in Information Entropy and SSIM compared to some existing methods. They attribute this to the trade-offs inherent in selective encryption and the specific statistical biases introduced by the chaotic map in certain coefficient domains. Future work is proposed to explore 2D extended Schaffer function maps and neural networks to further enhance adaptability and structural similarity.