Selective Chaotic Video Encryption for Versatile Video Compression H.266/ VVC Standard

In the current era, digital means, including international conferences over the Internet, distance education, and online gaming, have become widespread. However, this internet proliferation faces security challenges.

Encryption is the major method of ensuring data confidentiality and content integrity during data transmission or storage, particularly in open environments or untrusted networks, such as the Internet or wireless communications. Security of multimedia data is of significant concern in the majority of applications to enable secure storage and transmission of images and videos. The most straightforward solution to perform secure transmission of a video is to encrypt the whole video bit stream using a secure encryption algorithm such as Advanced Encryption Standard (AES) [1]. Video compression is a process used to reduce data size by removing redundant data in a smart and efficient manner. This reduces storage consumption and facilitates transportation across networks with minimal impact on video quality.

The H.266/Versatile Video Coding (VVC) is the latest generation of video compression algorithms, with the objective of offering a significant reduction in data size by up to 50% over the High Efficiency Video Coding H.265/HEVC at the same video quality. The VVC has sophisticated techniques that make this method suitable for applications requiring high quality and high compression efficiency, including 4K/8K virtual reality, cloud gaming, and live streaming [2].

Video content accounts for 82% of internet data, requiring large storage, processing time, and high bandwidth consumption. Traditional full encryption methods are insufficient due to data size and the slow full encryption of compressed video. Selective encryption of compressed video is considered the best solution to this problem; however, it faces gaps in both security and compression efficiency.

In recent years, the confidentiality of video data has become a challenging research topic. In 2013, Van Wallendael et al. [3] proposed encrypting specific elements of HEVC video using AES. Encryption showed a significant impact on quality when applied to high-motion videos. Some elements did not affect the compression rate, while others increased it. Ouamri et al. [4] discussed SE of HEVC videos by using AES-CBC as the encryption algorithm before the Binary Arithmetic Coding (BAC) stage. Encrypting only the suffixes without affecting the rest of the data may leave a fixed pattern that can be exploited to extract unencrypted information. Abu Taha et al. [5] introduced SE to protect Regions of Interest (ROI) within the video, using a chaotic system in the HEVC standard. The difference elements in Motion Vectors Difference (MVD), Motion Vector signs (MV Signs), Transformation Coefficients (TC), Transaction signs (TC Signs), and Intra Prediction Modes (IPM) are all encoded. Chen et al.  [6] proposed SE That relies on encrypting each slice using the RC4 algorithm. However, encrypting elements such as IPMs will increase the bit rate required for video transmission, which may affect compression efficiency. Farajallah et al. [7] proposed using VVC with SE. The encryption was done by a simple XOR operation between a binary sequence as a key and some selected parameters. This makes this system vulnerable to small attacks. Yu and Kim [8] proposed a method using the HEVC standard and ROI to encode motion vectors (MV), transform coefficients (TC), and intra prediction mode (IPM). At the same time, the visual distortion was more efficient and accurate than traditional encryption, but still, the use of HEVC may cause less compression efficiency as compared to VVC. Chen et al. [9] discussed a robust encryption method for compressed video using the VVC standard to secure video content. They used a 3-Dimensional Dynamical Chaos Model (3D-DCM) for encryption to perform encryption at the slice level, where each slice is encrypted independently. Encrypted information includes texture-related elements such as QTC transformation coefficients, MVD motion vectors, and IPM intra prediction.

Ibraheem et al. [10] proposed an algorithm based on the genetic algorithm, where they encoded Intra Prediction Modes, Transform Units (TUs), Quantization Parameters (QPs), and Entropy Coding Modes. The goal is to reduce and optimize encoding without affecting video quality. Shah et al. [11] proposed a new encryption algorithm, pixJS, based on chaotic systems. The algorithm is built on three main components: jumping process, logistic map, and linear feedback shift register. The algorithm exhibited high entropy and resistance to differential attacks. Usmani et al. [12] proposed a solution to improve the efficiency of 360-degree interactive video streaming. They used SE as a trade-off between security and processing speed. Although security was improved, this was at the expense of increased computation time.

In this paper, an SE is used with H.266/VVC compression standard, utilizing a two-dimensional sine cosine logistic map. The algorithm, based on chaotic systems, ensures high confidentiality and security while maintaining compression efficiency and video quality. The algorithm's performance was evaluated through security analysis, quality assessment, execution time, and metrics. The suggested technique combines the powerful VVC video compression and SE based on a highly efficient two-dimensional sine-cosine logistic map (2D-SCLM). The proposed chaotic SE is done directly inside the VVC bitstream, ensuring complete compatibility with the standard format and bit rate stability. The sensitive parameters in the video bitstream, such as I-frames and motion vector signs, are used, resulting in robust SE without impacting the decryption process. Unlike previous approaches that employ previous compression standards like AVC or HEVC, the proposed method uses VVC as a compression algorithm. Additionally, the use of such a chaotic map to encrypt part of the video will give increased security, enhanced key sensitivity, and greater resistance to statistical and differential attacks.

The rest of the paper is organized as follows: Section II provides an overview of the VVC standard structure and SE algorithm; Section III explains the chaotic map; Section IV shows the proposed algorithm. Section V gives the simulation results of the proposed algorithm. Some concluding remarks are provided in Section VI.

The proposed cryptographic key generation method uses a Two-Dimensional Sine Cosine Map (2D-SCLM). This combines sine, cosine, and logistic maps for (SE). This method has high randomness, high input sensitivity, and difficulty in making predictions about system behavior due to its non-linear nature and dependency on multiple dimensions.

The map possesses the feature of high potential for generating complex chaotic sequences and is therefore very suitable for generating random encryption keys in information security, especially selective video encryption.

The 2D-SCLM was chosen for its durability and also has many advantages, like high sensitivity to initial conditions and parameters, increased complexity and security through 2D coupling, uniform and pseudo-random distribution, efficient implementation, suitability for SE, improved statistical properties, and compatibility with video compression standards.

The map is defined by the following Eqs. (1) and (2) [20].

$x_{n+1}=sin \left(\pi x_n\right)+a\left(sin \left(\pi x_n\right)-sin ^2\left(\pi x_n\right)+sin \left(\pi y_n\right)\right)$           (1)

$\begin{gathered}y_{n+1}=cos \left(\pi y_n\right)+b\left(cos \left(\pi y_n\right)-cos ^2\left(\pi y_n\right)\right.  \left.+cos \left(\pi x_n\right)\right)\end{gathered}$           (2)

The control parameters $a$ &b (0, +∞), This increases the key size. The study used a Lyapunov exponent analysis to determine the ranges of a and b that lead to strong chaotic behavior. Results showed strong sensitivity to initial conditions at 1.7 and 0.67, and a deep chaotic regime at 0.67. Standard encryption metrics confirmed these parameters' excellent confusion and diffusion properties.

The 2D-SCLM has a wider chaotic range, larger Lyapunov exponent values, and complicated and non-linear chaotic dynamics. These make the generated keys highly random, highly sensitive to minimal changes in the initial keys (i.e., a trivial change resulted in completely different outputs). Complexity in the prediction of the system behavior, which strengthens the algorithm to withstand brute force and statistical attacks. Therefore, 2D-LSCM is employed as a chaotic key generator to selectively encrypt sensitive portions of the bit stream in the H.266/VVC compression standard, thereby achieving both security and efficiency. Figure 5 shows the attractors of 2D-SCLM.

05.png

Figure 5. The attractors of the 2D-SCLM are as follows: (a) the attractor described by parameters a = 0.01 and b = 1.7; (b) the attractor with parameters a = 0.67 and b = 1.7; and (c) the attractor with parameters a = 0.67 and b = 0.01 [20]

5.1 Video vision results

Here, the Quantization Parameter QP of the VVC encoder is set to 24, taking Akiyo, bus, and mobile videos as examples. The decoded results of some frames of the original video and the encrypted video are shown in Figure 7.

Here, the Quantization Parameter QP of the VVC encoder is set to 24, taking Akiyo, bus, and mobile videos as examples. The decoded results of some frames of the original video and the encrypted video are shown in Figure 8, where vision analysis shows low residual intelligibility, noise, and loss of main information of the frames.

08.png

Figure 8. Encrypted and decrypted frames for Akiyo, bus and mobile videos

5.2 Objective indicator results

Peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) are two objective indicators to further evaluate the quality of the encrypted frames. The PSNR is described as [7]

$\begin{aligned} \text {PSNR}=10 log \frac{1}{\frac{1}{M H X M V}-\sum_{i=1}^{M H} \sum_{j=1}^{m v}(X(i, j)-Y(i, j))^2}\end{aligned}$          (5)

where, MHxNV is the size of the video frame. X and Y represent the original frame and the encrypted frame, respectively. A smaller value of PSNR implies a higher level of encryption (less residual intelligibility). The SSIM is defined as [7]:

$SSIM=\frac{\left(2 \mu_X \mu_Y+\left(L K_1\right)^2\right)\left(2 \sigma_{X Y}+\left(L K_2\right)^2\right)}{\left(\mu_X^2+\mu_Y^2+\left(L K_1\right)^2\right)\left(\sigma_X^2+\sigma_Y^2+\left(L K_2\right)^2\right)}$           (6)

where, $\mu_x, \mu_y$ are the averages of X and Y and $\sigma_x^2, \sigma_y^2$ are variances of X and Y.L = 255, k₁ = 0.1, k₂ = 0.03, and SSIM $\in(0,1)$. The smallest values for SSIM refer to less residual intelligibility.

Table 1 shows PSNR and SSIM values for the eight selectively encrypted videos. This shows a significant decrease in both PSNR and SSIM values following the SE of video sequences. The PSNR metric is decreased due to distortion in the output image. The SSIM norm, quantifying structural similarity between original images and encrypted ones, also shows a drastic loss. The effect is weak for more stationary (less motion) videos like "akiyo" and "news". The low-motion videos like "akiyo" maintain quality better than high-motion videos. The results suggest a correlation between numerical loss of PSNR and structural similarity loss (SSIM), indicating that the degree of motion in the video type directly affects video quality after encoding. Of course, this is not related to SE, but it is related to the compression process.

Table 1. PSNR and SSIM values for the encrypted video

In this part of the study, the performance of the proposed system based on the VVC (H.266) standard, with other algorithms based on HEVC by Jovanović [21] and RSVE [6] is compared. This is due to the unavailability of papers that have tested VVC and contain results. The comparison includes measuring video quality using indicators such as PSNR and SSIM under various conditions, including different QP values. Tables 2 and 3 show the comparison between Jovanović and the RSVE HEVC SE technique and the proposed algorithm for PSNR and SSIM, respectively.

Table 2. PSNR comparison of the proposed algorithm with Jovanović and RSVE HEVC (SE) techniques

Table 3. SSIM comparison of proposed algorithm with Jovanović and RSVE HEVC (SE) techniques

5.3 Bit rate overhead results

The bitrate line shows the short-term impact of the proposed encryption algorithm on the size of the resulting information. The bit rate overhead is defined as in Eq. (7).

Bitrate

$\begin{gathered} {Overhead}= \left(\frac{ {Bitrate\ after\ encryption}- {Original\ Bitrate}}{ { Original\ Bitrate }}\right) \times 100\end{gathered}$           (7)

Table 4 shows the results of the bitrate overhead for compressed video, encrypted compressed video and bitrate overhead.

Table 4. The bitrate overhead for the original and encrypted video

5.4 Encryption time results

The proposed encryption algorithm's performance in terms of processing time for videos was evaluated. Table 5 shows that the average encoding time per frame ranges from 0.00505 seconds for "Carphone" to 0.01884 seconds for "bus". Videos with low complexity or fewer frames have higher efficiency, while more detailed or motion-heavy ones consume more time per frame. The algorithm operates within a fair timescale for real-world applications.

Table 5. Encryption time per frame for test videos

5.5 Information entropy results

Entropy measures randomness and data distribution in the video, and it is an excellent measure of encryption strength. The entropy must be higher after encryption, which means that there is an improvement in data randomness, and thus, predicting is more difficult [7].

$H(m)=\sum_{i=1}^{2^{L b}} P\left(m_i\right) log \frac{1}{P\left(m_i\right)}$                (8)

where, Lb is the number of bits per pixel and p(m_i) is the probability of pixel's level m_i.

Table 6. Entropy values for video sequences

Table 6 shows that all video sequences gained an increase in their entropy value after being selectively encrypted, indicating improved randomness and enhanced system reliability. This increase in entropy enhances the algorithm's ability to conceal the original.

5.6 Analysis of key space

The key space specifies the number of available keys that may be utilised to create various key streams. A larger key space makes the system more resistant to brute-force (exhaustive) attacks. The suggested approach uses the chaotic system's initial values (x₀, y₀) as the secret key.

Assuming both (x₀, y₀) are represented as 16-digit double-precision floating-point integers inside the region [0,1] [0,1], the number of potential values for each variable is about 10¹⁶. Therefore, the total key space may be estimated as follows, with 1000 iterations, as explained in Eq. (9).

$\begin{aligned} & s_k={Cardx}_0 \cdot { Card } y_0 \cdot{ Cardt } = 10^{16} \times 10^{16} \times 10^3=10^{35} \approx 2^{116}\end{aligned}$             (9)

where Card {} ⋅ represents the cardinality of a set.

Since the key space is approximately 2¹¹⁶, the system resists the Brute force attack.

5.7 Analysis of computation complexity

The encryption relied mainly on simple multiplication and addition operations on the encryption side, with multiplication and subtraction operations on the decryption side, which means the computational complexity is extremely low. Because of that, the proposed algorithm is high-speed and has minimal impact on the overall coding efficiency.

5.8. Key sensitivity analysis

A key sensitivity experiment was performed to evaluate the proposed chaotic encryption scheme's sensitivity to minor changes in the initial key settings. Two chaotic keys were produced with a 10-10 difference.

Key 1: initialized with x_0=0.1731234987

Key 2: initialized with x_0=0.1731234988

Key 1 is used to encrypt the bus video. The resulting encrypted video is decrypted with key 2. Figure 9 shows a frame of the encrypted video with key 1 and the decrypted video with key 2. Table 7 shows the key sensitivity result.

09.png

Figure 9. Key sensitivity result

Table 7. Key sensitivity results

This experiment highlights the significant key sensitivity of the chaotic encryption method. A slight modification in the initial condition produces significantly varied Decrypted outputs, an important feature against brute-force and differential assaults.

5.9 Objective indicator results for all frames

Figure 10 shows the objective indicator results (PSNR, SSIM, entropy of encrypted video, and encryption time per frame) for all frames of the Akiyo video. A sharp decline in PSNR and SSIM values is observed after the first frame, reflecting the impact of (SE) on image quality. Furthermore, the entropy remains consistently high (approximately 7.32), indicating strong randomness and effective encryption. Additionally, the encoding time stabilizes rapidly at a low value (around 0.01 to 0.02 seconds per frame), demonstrating the efficiency of the proposed method and its potential for real-time applications.

5.10 Statistical analysis

A test was conducted to confirm the effectiveness of the proposed method against statistical analysis attacks by examining the histogram of one of the videos. The histogram compares the distribution of pixels between the compressed video represented by the blue line and the compressed and encoded video represented by the red line, for each channel of the YUV color space.

The histogram in Figure 11 shows the strong statistical effect of the algorithm, where the blue distribution representing the original unencrypted compressed video displays clear and irregular patterns, with pixels concentrated in specific areas. These characteristics constitute a weakness, allowing an intruder to infer information or statistically attack the system. As for the compressed and encrypted video, represented by the blue histogram, it appeared almost flat and uniform across all color channels, which completely proves the randomness of the data.

10.png

Figure 10. Objective indicator results for all frames of the Akiyo video

011.png

Figure 11. Histogram analysis