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

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

1

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

arXiv:2608.22573v1 [cs.CR] 23 Aug 2026

Liang-Wei Li, Chung-Nan Lee, Member, IEEE, Kishu Gupta, Member, IEEE, Huei-Fang Yang, Member, IEEE, Ashutosh Kumar Singh, Senior member, IEEE

Abstract—In today’s digital landscape, high-efficiency video coding (H.265/HEVC) has emerged as the most widely used video coding standard, employing selective encryption schemes to protect the privacy of video content while maintaining efficient compression performance. However, existing coefficient scrambling methods impose a significant computational load, leading to increased bit rate overhead due to encryption, longer execution times, and insufficient safety measures. To address these issues, a new coefficient scrambling scheme based on chaotic maps is proposed. This approach leverages the pseudorandomness, ergodicity, and sensitivity to initial conditions inherent in chaotic maps to generate highly unpredictable coefficient distributions, thereby strengthening security while preserving low complexity. Unlike conventional scrambling, chaotic maps ensure minimal correlation between encrypted coefficients, enhancing resistance against statistical and differential attacks. Additionally, the scrambling conditions are specifically designed to minimize the impact on the bit rate overhead. Furthermore, when combined with syntax element encryption (SEC), which includes motion vector difference (MVD), quantized transform coefficients (QTC), and luma intraprediction mode (Luma IPM), this method effectively distorts video content. The proposed scheme operates synchronously with slices, ensuring that the decryption of video content remains intact even if some slices are lost. Additionally, a random sequence generated by AES-CTR is incorporated with the H.265 encoded stream to protect against chosen-plaintext attacks. The experimental results indicate that this scheme features high security, compliance with format standards, fast execution times, synchronous updates with slices, and resilience against common attacks, all while achieving a reduced bit rate overhead of 45.13% with a lowered average execution time overhead of 1.91%. Index Terms—H.265, selective encryption, syntax element encryption, chaotic map, coefficient scrambling, AES-CTR

I. I NTRODUCTION Manuscript received Received 15 September 2025; revised 9 October 2025; accepted 19 October 2025. Date of publication 23 October 2025; date of current version 7 April 2026. This work was supported in part by the National Science and Technology Council, Taiwan, under Grant NSTC 1142634-F-110-001-MBK, Grant 114-2640-E-110-004, and Grant 112-2221-E110-047-MY3. This article was recommended by Associate Editor G. Xu. (Corresponding author: Kishu Gupta.) Liang-Wei Li, Chung-Nan Lee, and Huei-Fang Yang are with the Department of Computer Science and Engineering, National Sun Yat-sen University, Kaohsiung, 80424, Taiwan. (e-mail: zaq871210@gmail.com, cnlee@mail.cse.nsysu.edu.tw, hfyang@mail.cse.nsysu.edu.tw). Kishu Gupta, is with Department of Computer Science and Engineering, National Sun Yat-sen University, Kaohsiung, 804, Taiwan, and also with Department of Computer Science, the VIZJA University, 01-043 Warsaw, Poland. (e-mail: kishuguptares@gmail.com). Ashutosh Kumar Singh, is with Department of Computer Science and Engineering, Indian Institute of Information Technology Bhopal, Bhopal 462003, India, and also with Department of Computer Science, the VIZJA University, 01-043 Warsaw, Poland. (e-mail: ashutosh@iiitbhopal.ac.in). Digital Object Identifier 10.1109/TCSVT.2025.3625077

W

ORLDWIDE rapid growth in video applications such as video-on-demand (VOD), video conferencing, online medical consultations, online courses, and video surveillance has resulted in massive volumes of streaming data. The sensitive information within these videos is at risk of leakage during transmission and storage due to the limited capabilities of streaming channels and cloud service providers. To safeguard video content from privacy erosion and copyright violations caused by data breaches, multiple solutions have been proposed [1]–[4]. The selective encryption algorithm (SEA) encrypts parts of the H.265 encoded stream (syntax elements) to induce significant distortion while reducing computational complexity; however, it remains vulnerable to leakage of highly sensitive data due to partial content protection [1]. A recent memristive Rulkov neuron–based approach introduced an innovative chaos-driven model characterized by strong nonlinearity, a large keyspace, and efficient segmentwise encryption; however, its limited cryptographic validation, finite-precision issues, and potential data leakage constrain its robustness [5]. Another notable contribution provided a comprehensive review of chaos-based video encryption techniques and current research trends, yet it lacked experimental validation and relied heavily on previously reported findings [6]. Furthermore, critical factors such as encryption strength, computational efficiency, format compliance, bit rate overhead, and perceptual video quality must be carefully balanced to ensure secure, correct decoding and seamless playback. Inadequate consideration of these factors may lead to excessive bit rate escalation or playback distortion, undermining the overall effectiveness of video encryption schemes [7]. In this context, this paper proposes a novel approach to furnish video content security and privacy by incorporating encryption into selective video content via chaotic maporiented coefficient scrambling. It only encrypts five syntax elements, including motion vector difference (MVD) information (mvd sign flag, abs mvd minus2 suffix), quantized transform coefficient (QTC, coefficient for short) information (coeff sign flag, coeff abs level remaining suffix) and luma intra prediction mode (Luma IPM) information. Furthermore, at the same time, a coefficient scrambling scheme based on chaotic maps is devised to disturb the coefficient distribution effectively. Additionally, the proposed scheme designs effective coefficient scrambling conditions ushering toward the reduced bit rate overhead rendered by coefficient scrambling. This proposed scheme uses the AES-CTR to generate a random sequence, and the encryption operation is associated with the random sequence. The generated random sequence

1051-8215 © 2025 IEEE. All rights reserved, including rights for text and data mining, and training of artificial intelligence and similar technologies. Personal use is permitted, but republication/redistribution requires IEEE permission. See https://www.ieee.org/publications/rights/index.html for more information. This article has been published in IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY © 2025 IEEE.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

is synchronized with the slice to ensure that the loss of some slices does not affect the decryption of the video content. In addition, this paper uses the SHA-384 hash value of the slice header to update the initial key and initial vector to resist the chosen plaintext attack.

2

Update context modeling based on binary value

Non-Binary Syntax Element Binarization Syntax Element

Binary Syntax Element

bin string

Context Modeling

Regular Encoder Bypass Encoder

H.265 Encoded Stream

Binary Arithmetic Coder

A. Key Contributions The main contributions of this paper are as follows: • This paper presents a novel, robust H.265/HEVC video encryption scheme that aims to ensure video privacy and security. It analyzes the relationship between selective syntax elements and encryption, leveraging chaotic mapbased coefficient scrambling. • This scheme emphasizes the impact of encryption via selective element encryption, which involves three types of information: MVD, QTC, and Luma IPM (described in Section III-C), to effectively distort the video content to ensure compliance with format standards and contributes to the results, including bitrate reduction, execution time overhead reduction, and resistance to substitution attack. • An efficient coefficient scrambling scheme using chaotic maps, which leverages the pseudorandomness of these maps to effectively disrupt the distribution of coefficients, is presented. • The scheme carefully establishes the conditions for coefficient scrambling to protect video edges and minimize the bit rate overhead significantly caused by coefficient scrambling. • A comparative analysis, both theoretical and experimental, demonstrated significant improvements in various performance metrics compared with state-of-the-art methods. Paper Organization: Section II highlights state-of-the-art approaches for Context-based Adaptive Binary Arithmetic Coding (CABAC), the Selective Encryption Algorithm (SEA), and a chaotic map. Section III outlines the detailed function of the proposed model. Section V offers an operational summary of the proposed approach along with the performance metrics. Section VI entails performance evaluation followed by discussion and limitations of the proposed approach in Section VII. Conclusive remarks and the future scope of the proposed work are presented in Section VIII. II. R ELATED W ORK A. Context-based Adaptive Binary Arithmetic Coding The input syntax element needs to be converted into a Bin string through binarization, as illustrated in Fig. 1. However, if the syntax element itself is represented in binary, binarization can be ignored. Binarization methods include unary code, fixed length code (FL code), truncated rice code (TR code), exponential golomb code (Exp-Golomb code), etc., followed by encoding the Bin string. In the regular coding mode, an appropriate probabilistic model is first assigned to each bin string currently input into the context modeler based on the previously encoded bin string. Furthermore, the bin string and the newly configured probability model are input into the binary arithmetic coder

Fig. 1: The overall architecture of CABAC. for encoding. After encoding, the context modeling needs to be updated adaptively based on the output encoding stream. In bypass coding mode, encoding is performed directly with equal probability (0 and 1 each account for half the probability). B. Selective Encryption Algorithm for H.265/HEVC The selective encryption algorithm (SEA) emphasizes security, execution time, format compliance, and bit rate overhead. Format compliance ensures that an encrypted video must be decoded by the decoder and that the video plays successfully, regardless of whether the decryption operation is performed. On the other hand, an encrypted video without decryption will play but appear distorted. A few specific reasons for increasing the bit rate overhead are as follows: 1) When syntax elements are encrypted in regular coding mode, modifications to the encoded data may alter the statistics of the encoded stream, and 2) directly encrypting syntax elements via some binarization method (unary code, TR code, Exp-Golomb code, etc.) can revise the length of the bin string. These prerequisites result in an increased bit rate for the encrypted video. Lee et al. [2] proposed SEA based on the start code. However, according to Lee et al. [8], the encrypted start code can be cracked through a ciphertext-only attack, which is not suitable for practical applications. Wallendael et al. [9] studied the encryption of many syntax elements, and the results revealed that encrypting the MVD sign and the QTC sign can cause large distortions and a constant bit rate. Peng et al. [10] proposed an H.265 adjustable SE scheme based on chroma IPM and edge coefficient scrambling. The Enc scheme encrypts multiple syntax elements and still reveals some edge information, so the Scr scheme, which adds an edge coefficient scrambling scheme but reduces the bit rate overhead, deserves further study. Sallam et al. [11] used the RC6 encryption algorithm to encrypt syntax elements in different operating modes and reported that the best effect was achieved in the CFB operating mode. Farajallah et al. [3] proposed two encryption schemes based on the tile concept, locating the region of interest (ROI) in the tile. Zhang et al. [12] proposed an SE scheme based on the improved CABAC method to optimize the regular coding mode. Wen et al. [13] proposed three levels of H.265 SE schemes, and different levels of encrypted videos are used in different scenarios. Yang et al. [14] proposed extracting the subencoding stream from the encoded video and encrypting it to avoid the problem of reencoding for the encryption operation. Xu et al. [15] proposed an HEVC efficient exchange encryption and data hiding scheme. Boyadjis et al. [16] proposed the H.264 and

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

H.265 luma IPM encryption methods, showing that encrypted luma IPM is sufficiently destructive to I frame content. Chen et al. [17] proposed an RSVE scheme. The generated random sequence is synchronized with the slice, and the RC4 key is updated through the slice header. The syntax elements are encrypted for MVD, QTC, and IPM information, and combined with coefficient scrambling, the two-round shift method is used to perturb the coefficients. The results show that the RSVE scheme has the characteristics of high security, resistance to common attacks, fast execution time, etc., but reducing the bit rate overhead deserves further study. Although the RSVE scheme proposed by Chen et al. [17] is relatively robust, it lacks an improved overhead bit rate. This paper exploits the RSVE scheme as a benchmark to propose a new coefficient scrambling scheme to effectively reduce the bit rate overhead caused by coefficient scrambling. Additionally, the proposed coefficient scrambling scheme applies to the Scr scheme of Peng et al. [10] and the tunable selective encryption scheme of Sallam et al. [18]. C. Chaotic Maps for Pseudorandom Values Chaotic maps continually and iteratively generate pseudorandom values through a set of mathematical models. The generated values have characteristics such as ergodicity, randomness, nonperiodicity, and high sensitivity to initial conditions. Chaotic maps are commonly used to create random sequences for video encryption and can offer a large key space. Zhang et al. [19] employed a hyperchaotic Lorenz system to generate random sequences. Sallam et al. [20] used random bits generated by a chaotic logistic map (CLM) to encrypt the MVD sign and the QTC sign. Liu et al. [21] exploited an integer dynamic coupling tent map to generate random sequences and supported multicore parallelization operations to increase the speed of generating random sequences. Ye et al. [4] improved the ICMIC system, combined it with CML, and proposed L-ICMIC-CML to generate random sequences. Moreover, Lu et al. [22] provided an efficient scheme to encrypt image leveraging LSS Chatoic Map and Single S-Box.

3

H.265 Encoder Transform and Quantization

-

Frame

Inverse Transform and Inverse Quantization

+ Deblocking and SAO Filter Intra/ Inter Prediction

Frame Buffer

Fig. 2: The workflow of the proposed scheme. Syntax Element

Binary Selective Bin Conversion String Encryption

Binary Arithmetic Code

Encrypted H.265 Bitstream

Fig. 3: CABAC and SEE module.

slice header and slice data. First, the initial key and initial vector are updated via the slice header. Then, AES-CTR is used to create a random sequence for coefficient scrambling and encryption of syntax elements. After that, coefficient scrambling is performed for the QTC, and important syntax elements in the slice data are selected for encryption. Finally, the encrypted slice data and slice header are combined to form an encrypted slice. Original Slice

Slice Header

Slice Data

Random Sequence Generation

Coefficient Scrambling

Syntax Element Encryption Encrypted Slice Data

III. P ROPOSED S CHEME The proposed scheme for encrypting H.265/HEVC video is driven by chaotic map-based coefficient scrambling and syntax element encryption. It comprises three subparts, as illustrated in Fig. 2, Fig. 3, and Fig. 4, collaboratively. In the H.265 encoding process depicted in Fig. 2, data generated from the intra/inter prediction, transform and quantization, and deblocking and SAO filter modules are input into the H.265 encoder. The encoder then uses CABAC to encrypt certain syntax elements. Once the encoding is completed, an encrypted H.265 stream is produced. The internal architecture of CABAC and the selective encryption module is illustrated in Fig. 3. Any data input to CABAC is referred to as a syntax element. The encryption process for certain syntax elements begins after they have been converted into a binary string through binary conversion. Furthermore, the encryption process of the proposed scheme is illustrated in Fig. 4. The original slice is divided into a

Encrypted H.265 Bitstream

CABAC and Selective Encryption

Encrypted Slice

Fig. 4: Encryption process.

A. Random Sequence Generation The AES-CTR is used to create a random sequence considering the initial key size of 256 bits and the initial vector size of 128 bits. The process for generating slice key and slice IV is explained in the following steps: Step 1: Obtain the slice header and input it into SHA-384 to generate a 384-bit slice header hash. Step 2: Perform the XOR operation on the first 256 bits of the slice header hash and the initial key to obtain slice key and perform the XOR operation on the last 128 bits of the

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

slice header hash and the initial vector to obtain slice IV as computed in Eqs. (1) and (2), respectively. slice key = slice header hash ⊕ initial key

(1)

slice IV = slice header hash ⊕ initial vector

(2)

The process is repeated each time the encoder encodes a new slice to update slice key and slice IV. The newly generated slice key and slice IV are then used as inputs to the AESCTR to produce a set of random sequences with a length of 64 bits. Additionally, the slice IV is incremented to generate a different random sequence for the next time.

4

and exclude the existence of the first coefficient greater than 1 in the front section of C as 2. The range for the front section is 1 ∼ min(n, 8); if it exists, set F = 1. After that, the coefficients are eliminated; if N ≤ 1, end the coefficient scrambling process, and jump to step 12. • Step 5: Now, the pseudo-random value R reads N pseudorandom values R′ from index B. ′ • Step 6: Sort R to generate the sorted index D. The process for sorting index generation is illustrated in Fig. 6. 0

1

2

3

4

0.7319

0.5843

0.9143

0.5487

0.1873

4

3

1

0

2

0.1873

0.5487

0.5843

0.7319

0.9143

4

3

1

0

2

B. Coefficient Scrambling Sort

The coefficient scrambling process comprises a series of steps, as depicted in Fig. 5, and is explained below.

Generate a set of pseudorandom values using chaotic map

Read first and second bytes of slice key as starting index

of

Enter set of coefficients and number of coefficients

Yes Next coefficient scrambling is new slice

Coefficients scrambling condition (step 4) reads pseudorandom values starting at index

Fig. 6: Sorted index generation. •

No

Shift coefficient to right according to

Sort to generate sorted index

Step 7: Determine the number of coefficient scrambles S by using Eq. (4), which needs to be associated with the random sequence value Si . S=

mod (Si , N − 1) + 2; 0 ≤ Si ≤ 15

(4)

Step 8: Update B through S; here, set B = B + S. Step 9: D extracts S from the beginning as coefficient scrambling index values X. • Step 10: If F= 0, check the exclusion coefficient scrambling index value condition. The condition is whether C and X exist when C[X[i − 1]] = 2 and X[i] = 0. If this occurs, exclude X[i], S− 1. After the coefficient scrambling index values are eliminated, if S ≤ 1, terminate the coefficient scrambling process and jump to step 12. • Step 11: Shift the coefficient to the right according to X. The process of coefficient perturbation is depicted in Fig. 7. •

Exclude coefficient scrambling index value condition (step 10)

extracts from beginning as coefficient scrambling index values

Update through

Read random sequence to determine number of scrambles

Fig. 5: Chaotic map-based coefficient scrambling process. Step 1: Generate a set of pseudorandom values R using chatoic map through Eq. (3). The system combines the nonlinear dynamics of the logistic and sine maps to enhance ergodicity, sensitivity to initial conditions, and statistical randomness while avoiding the computational burden of higher-dimensional chaotic systems. The logistic-sine low-dimensional discrete chaotic system proposed in [21] is used to generate pseudorandom values to avoid excessive execution time. 4µ 9−µ x(n + 1) = ∗ x(n) ∗ (1 − x(n)) + ∗ sin[π ∗ x(n)] 9 9 (3) where 0 ≤ µ ≤ 9 and where x(n) is a function used to generate pseudorandom values ∈ R. It is assumed that µ = 7.3, x(0) = 0.9324, and it iterates 65535 times, resulting in a total of 65536 pseudorandom values. • Step 2: Determine the starting index B of R by selecting the first and second bytes of slice key. • Step 3: Specify the coefficient group C and the total number of coefficients N. If N ≤ 1, then end the coefficient scrambling process and jump to step 12 directly. • Step 4: Check the coefficient scrambling conditions. The conditions exclude the existence of coefficients in C as 1 •

•

0

1

2

3

4

3

2

2

3

4

3

3

4

2

4

3

1

2

Fig. 7: Coefficient perturbation. Step 12: If the next coefficient scrambling is a new slice, return to step 2; otherwise, return to step 3. Compared with the coefficient scrambling scheme proposed by related research, the edge coefficient scrambling scheme proposed by the Scr scheme of [10] uses the exchange value method to perturb the coefficients. However, the proposed scheme is not associated with a random sequence and uses only the rand function to determine the exchange object, which •

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

results in extended security. The coefficient scrambling scheme of [17] uses a two-round shift for perturbation, as illustrated in Fig. 8.

5

encryption is allowed, as does the number of encryption bits L. If encryption is allowed, the remaining Coeff Suffix is encrypted via Eq. (11). en remaining Coef f Suf f ix = remaining Coef f Suf f ix ⊕ Si ; 0 ≤ Si ≤ 2L

0

1

2

3

4

5

6

7

8

0

8

1

3

4

2

3) Luma IPM Information: The Luma IPM comprises 35 prediction modes, including planar mode, DC mode, and 33 angle modes. Encryption luma IPM is performed via Eq. (12).

9

shift

step=3 6

7

5

(11)

en luma IP M =

9

mod (luma IP M + Si , 35); 0 ≤ Si ≤ 63 Fig. 8: One-round shift. If the step is larger, the amplitude of the disturbance is smaller, and for a smaller step, it is equivalent to shifting all the coefficients to the right, and the degree of disturbance is still not good. Therefore, the coefficient scrambling scheme proposed by [17] is still limited in the distribution of coefficient perturbations. The scheme proposed in this paper successfully furnishes a more random coefficient perturbation distribution.

(12)

Directly encrypting luma IPM will cause format-compliant failure; therefore, luma IPM and en luma IPM are first recorded, followed by three additional actions. 1) To determine the three most likely candidate prediction modes of adjacent PUs, en luma IPM is used to obtain the luma IPM of adjacent PUs. 2) The coefficient scan mode is associated with luma IPM [16] and is determined via en luma IPM. 3) Chroma IPM is also associated with luma IPM, as depicted in Table I.

C. Syntax Element Encryption 1) MVD Information: The encryption operations MVD Hor Sign and MVD Ver Sign are performed via Eqs. (5) and (6), respectively. en M V D Hor Sign =

(5)

M V D Hor Sign ⊕ Si ; 0 ≤ Si ≤ 1 en M V D V er Sign =

(6)

M V D V er Sign ⊕ Si ; 0 ≤ Si ≤ 1

The MVD size consists of the base level and the remaining level. The structure of the remaining level is the [N-1 zeros prefix][1][N bits suffix]. The remaining MVD Hor Suffix and remaining MVD Ver Suffix are encrypted via Eqs. (7) and (8), respectively. en remaining M V D Hor Suf f ix = remaining M V D Hor Suf f ix ⊕ Si ; 0 ≤ Si ≤ 2N en remaining M V D V er Suf f ix = N

remaining M V D V er Suf f ix ⊕ Si ; 0 ≤ Si ≤ 2

(7)

(9)

The QTC size consists of the base level and the remaining level. The structure of the remaining level is [M-1 ones prefix][0][M bits suffix]. The rice parameter r of the TR code is computed via Eq. (10). r

r = min(4, r + 1) if QT C size > 3 × 2

Chroma IPM ID

Chroma prediction mode

0 (Planar) 26 (Vertical) 10 (Horizontal) 1 (DC) 34

0 (Planar) 4 1 2 3 0

Corresponding luma prediction mode 26 10 1 (Vertical) (Horizontal) (DC) 0 0 0 4 1 1 2 4 2 3 3 4 1 2 3

34 0 1 2 3 4

If the corresponding luma IPM is encrypted and the chroma IPM ID changes, a different chroma IPM is obtained during decoding. Therefore, before encoding chroma IPM, we must first determine whether chroma IPM needs to be adjusted; as shown in Algorithm 1. Furthermore, when the chroma IPM ID is determined, the corresponding luma IPM uses en luma IPM. Additionally, the coefficient scan mode is also associated with chroma IPM, and the adjusted chroma IPM is used to make final decisions. IV. T HEORETICAL A NALYSIS

(8)

2) QTC Information: The encryption coeff Sign is performed via Eq. (9). en coef f Sign = coef f Sign ⊕ Si ; 0 ≤ Si ≤ 1

TABLE I: Association of chroma IPM and luma IPM

(10)

Encryption of the remaining Coeff Suffix may change the rice parameter, causing format-compliant failure. Therefore, the remaining Coeff Suffix requires certain conditions before

A. Formal Theoretical Proof to Resist Chosen-Plaintext Attacks Theorem 1. The proposed Syntax Element Encryption for H.265/HEVC using Chaotic Map-Based Coefficient Scrambling Scheme is secure against Chosen-Plaintext Attack (CPA) under the assumption that the underlying chaotic map is computationally indistinguishable from a random permutation and the SHA-384 based key update process behaves as a pseudo-random function (PRF). Proof. Under the Chaotic Map Pseudorandomness Assumption (CMPA) the output sequence S ( initialized with a secret key K) is indistinguishable from a truly random permutation without knowledge of K. Assume for contradiction there exists a probabilistic polynomial-time (PPT) adversary A breaks the CPA security of our scheme with non-negligible advantage ϵ.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

We construct a simulator B that uses A to distinguish between a real chaotic map and a random permutation, thereby breaking CMPA or PRF assumptions: • B receives a real chaotic permutation derived from a secret key K, or a truly random permutation. • A selects two plaintext coefficient sets C0 and C1 . • B selects a random bit b ∈ {0, 1}, scrambles Cb using the permutation provided by C, and returns the ciphertext Cb′ to A. Here, C is the plaintext syntax element set being scrambled and encrypted. ′ • If b = b, then B guesses that the permutation is real chaotic map; otherwise, random. If A has advantage ϵ, so does B, violating CMPA or PRF assumptions. Since these primitives are secure by hypothesis, ϵ must be negligible. Thus, the proposed scheme resists CPA.

B. Chaotic System Dynamics Analysis Theorem 2. The Logistic-Sine Chaotic Map (LSCM) used in the proposed Syntax Element Encryption for H.265/HEVC exhibits properties of a robust chaotic system namely, parameter sensitivity, chaotic region validity, and positive Lyapunov exponent ensuring the unpredictability and security of the generated pseudorandom sequences. Proof. Consider the LSCM, as defined in Eq. (3), where x(n) ∈ (0, 1) is the chaotic state at iteration n, and µ ∈ [0, 9] is the control parameter. • Parameter Sensitivity: The LSCM exhibits strong sensitivity to initial conditions and parameters. For two nearby initial states x0 and x0 + δ, the divergence satisfies below condition: |xn − x′n | ≈ eλn |δ|, where λ denotes the Lyapunov exponent. A positive λ implies that small differences in x0 or µ result in exponentially divergent sequences, ensuring unpredictability of the generated pseudorandom numbers. • Chaotic Region Validation: To ensure fully chaotic behavior, the control parameter µ is selected within a specific range that maintains non-periodicity and complexity of the output. Accordingly, the system exhibits strong chaotic characteristics for µ ∈ [6, 9]. Additionally, the initial condition x0 ∈ (0, 1) \ {0, 1} prevents trivial orbits and degeneration into fixed points or short cycles. • Lyapunov Exponent Analysis: The system’s chaoticity is quantified by the Lyapunov exponent defined as

This analysis verifies that the LSCM provides high sensitivity, validated chaotic behavior, and positive Lyapunov exponent ensuring secure pseudorandom sequence generation for coefficient scrambling in the proposed H.265/HEVC syntax element encryption scheme. C. Theoretical Proof for Information Entropy Analysis Theorem 3. The proposed Chaotic Map-Based Coefficient Scrambling Scheme for H.265/HEVC significantly increases the information entropy of syntax elements, resulting in enhanced randomness and resistance against statistical and entropy-based attacks. Proof. For a discrete random variable X with possible outcomes {x1 , x2 , . . . , xn } and corresponding probabilities P (X) = {p1 , p2 , . . . , pn }, the Shannon entropy H(X) is computed using Eq. (22). • Pre-Encryption Entropy: Let C = {c1 , c2 , . . . , cn } represent the set of residual coefficients before encryption (i.e. original), following a Laplacian-like distribution:   |c| 1 exp − P (c) = 2b b which implies a higher probability of small coefficient values (0, ±1), and lower probabilities for larger values. This distribution is peaked around zero, resulting in lower entropy. Therefore, the entropy of C satisfies: H(C) < Hmax = log2 n For 8-bit coefficients (n = 256), Hmax = 8 bits/symbol. Empirical HEVC data yields H(C) ≈ 5.5 bits/symbol i.e. commonly in range of 5 to 6 bits/symbol. • Post-Encryption Entropy: After applying the chaotic mapbased scrambling, two possible scenarios are: – Permutation-only Scrambling: Histogram remains unchanged; Entropy H(C ′ ) ≈ H(C); Spatial correlation destroyed. – Chaotic Value Masking: Each coefficient is modified as: c′i = ci ⊕ ki where ki is drawn from a pseudo-random sequence generated by the Logistic-Sine Chaotic Map (LSCM). This produces an approximately uniform distribution: 1 P ′ (x) ≈ b , b = 8 2 yielding: H(C ′ ) ≈ 8 bits/symbol.

n−1

1X ln |f ′ (xi )| , n→∞ n i=0

λ = lim

where the derivative f ′ (x) of the map (in Eq. (3)) is defined as: 4µ 9−µ f ′ (x) = · (1 − 2x) + · π · cos(πx). 9 9 For µ ∈ [6, 9] within the chaotic region, numerical evaluation yields λ > 0, confirming the system’s chaotic nature and suitability for cryptographic applications.

6

•

Theoretical Entropy Gain: The entropy gain ∆H satisfies: ∆H = H(C ′ ) − H(C) ≈ 8.0 − 5.5 = 2.5 bits/symbol,

demonstrating a significant increase in randomness and unpredictability. Due to this entropy enhancement, the encrypted coefficients closely resemble uniformly random data, reducing the risk of entropy-based and statistical attacks. Thus, the scheme improves security from an information-theoretic perspective.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

7

D. Security Proof

B. Performance Metrics

Theorem 4. The proposed Syntax Element Encryption for H.265/HEVC using Logistic-Sine Chaotic Map (LSCM) and SHA-384 key update mechanism ensures information-theoretic security, including large key space, high entropy, strong confusion and diffusion, and resistance against Chosen-Plaintext Attacks (CPA).

The two indicators peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) are computed for further evaluation. The PSNR value is computed via Eq. (13).

Proof. Key Space: With x0 ∈ (0, 1) and µ ∈ [6, 9] at 10−15 precision, and SHA-384 key updates (2384 possibilities), the total key space exceeds 2300 , far above the 2128 security benchmark. Entropy: Encrypted syntax elements approach a uniform distribution: H(X) ≈ log2 N, ensuring high unpredictability and randomness. Confusion and Diffusion: LSCM provides nonlinearity (confusion), while dynamic SHA-384 key updates propagate changes across syntax elements (diffusion), ensuring the avalanche effect. CPA Resistance: Assuming LSCM outputs are computationally indistinguishable from random and SHA-384 behaves as a pseudo-random function, the scheme achieves security. Mutual Information:

(2N − 1)2 (13) M SE where N represents the grayscale level of the image. This scheme assumes that N = 8, which is an 8-bit grayscale image. The mean square error (MSE) is expressed by Eq. (14). 2 H X W  X 1 M SE = (14) X(i, j) − Y(i, j) H × W i=1 j=1 P SN R(X, Y) = 10 log10

where H and W define the height of the image and the width of the image, respectively. The PSNR is measured in dB. A lower PSNR value indicates a lower similarity between the two frames. The SSIM values are computed via Eq. (15). SSIM (X, Y) =

(2µX µY + C1 )(2σXY + C2 ) 2 (µX + µ2Y + C1 )(σX2 + σY2 + C2 )

where C1 = (K1 L)2 , C2 = (K2 L)2 , the default value of K1 is 0.01, the default value of K2 is 0.03, and L denotes the maximum grayscale value of the image; for an 8-bit grayscale image, L is 255. µX , σX , and σXY represent the mean, standard deviation, and covariance, respectively, as computed via Eqs. (16-18).

I(M ; C) = H(M ) − H(M |C) ≈ 0, implying ciphertext reveals negligible information about plaintext. The scheme satisfies key cryptographic criteria: large key space, high entropy, strong confusion and diffusion, and CPA resistance.

V. O PERATIONAL D ESIGN AND P ERFORMANCE M ETRICS A. Operational Design The operational summary for the proposed scheme is described in Algorithm 1. Algorithm 1: Proposed Scheme: Operational summary Input chroma IPM, luma IPM, and en luma IPM Output updated chroma IPM 3 Begin: 4 if chroma IPM=34 and luma IPM ̸= en luma IPM then 5 chroma IPM=luma IPM 1

2

6 7 8 9

else if chroma IPM = en luma IPM then chroma IPM=34 else no action

end if 11 End 10

(15)

m

1 X Xi m i=1 s 2 m X Xi − µ X

µX =

1 m − 1 i=1   m  1 X Xi − µ X σXY = Yi − µ Y m − 1 i=1 σX =

(16)

(17) (18)

where m is the total number of pixels in the image. The computations for µY and σY are similar to those for µX and σX . The SSIM value ranges between [0, 1]. A smaller value indicates a lower similarity between the two frames. Another important metric, the edge difference ratio (EDR) indicator required for performance evaluation, is computed via Eq. (19). Pω i=1 |EX (i) − EY (i)| (19) EDR(X, Y) = Pω i=1 |EX (i) + EY (i)| where ω represents the total number of pixels in the edge frame. The value of the EDR ranges from 0 to 1. A higher value indicates a greater difference between the two edge frames. The encryption operation may increase the file size of the encrypted video, so the bit rate overhead is also one of the considerations when choosing to encrypt the video. The formula for bit rate overhead is expressed in Eq. (20). ES − OS × 100% (20) OS where ES is the file size of the encrypted video and where OS is the file size of the original video. The unit of bitrate BRO =

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

overhead is the percentage (%), and the ideal value is 0%, which means a constant bitrate. The average Hamming distance (HDavg ) to calculate the difference between two sets of random sequences and to evaluate the resistance to chosen plaintext attacks is computed via Eq. (21). PKL Sx (i) ⊕ Sy (i) HDavg = i=1 (21) KL where KL represents the minimum length of the two sets of random sequences and where Sx (i) and Sy (i) represent the ith values of the first and second sets of random sequences, respectively. The value of HDavg is between 0 and 1. The closer its value is to 0.5, the greater the difference between the two sets of random sequences is, and it can effectively resist chosen plaintext attacks. The information entropy is computed via Eq. (22). H(X) = −

N X

P (Xi ) × log2 P (Xi )

(22)

i=0

where N represents the maximum grayscale value of the image. For an 8-bit grayscale image, N is 255. P (Xi ) represents the probability of occurrence of frame X when the pixel value is i. For an 8-bit grayscale image, the maximum value of information entropy is 8. A value of information entropy closer to 8 reflects a higher degree of randomness. Furthermore, widely used to test the robustness of image cryptosystems against differential cryptanalysis, the number of pixels changing rate (NPCR) and unified average change intensity (UACI) are computed via Eqs. (23–25). P u,v O(u, v) × 100% (23) N P CR = H×W ( 0, if C1 (u, v) = C2 (u, v) O(u, v) = (24) 1, if C1 (u, v) ̸= C2 (u, v) 1 X C1 (u, v) − C2 (u, v)  U ACI = × 100% (25) H × W u,v 2L − 1 where C1 (u, v) and C2 (u, v) represent the pixel values of two cipher frames at position (u, v). The expected values of the NPCR and UACI are computed via Eqs. (26) and (27), respectively. N P CR† = (1 − 2−L ) × 100% P2L −1 u(u + 1) 1 † U ACI = 2L u=1L × 100% 2 2 −1 where L =8 is the length of the pixel values.

(26)

8

key is set to {0x04, 0xB0, 0x21, 0x90, 0xCB, 0xA6, 0xAF, 0xF6, 0x94, 0xFD, 0xAA, 0x9B, 0xEB, 0xEE, 0xA6, 0xCD, 0xF1, 0xF2, 0xF1, 0xBC, 0xCE, 0x17, 0x45, 0xE2, 0x39, 0x57, 0x68, 0x8A, 0xCB, 0x8A, 0x21, 0x5A}, a total of 256 bits, the initial vector is set to {0x17, 0x3B, 0xFC, 0xD8, 0x7B, 0x33, 0x0D, 0x30, 0x0D, 0xDB, 0x12, 0x65, 0xAF, 0xA9, 0x47, 0x52}, with a total of 128 bits. TABLE II: Experimental environment configuration Item Operating System CPU RAM Software Programming Language Library H.265 Codec

Specification Microsoft Windows 10 64bit Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz 24.0 GB Microsoft Visual Studio 2019, MATLAB C++ Crypto++ HM 17.0

The simulation and experiments employed a 12-video sequence with resolutions ranging from 352×288 to 2560×1600 [24], [25], as shown in Table III, and were compared with the robust RSVE scheme [17]. TABLE III: Video sequence Video Mobile Foreman BlowingBubbles RaceHorses PartyScene BasketballDrillText Johnny KristenAndSara BasketballDrive BQTerrace Traffic PeopleOnStreet

Resolution 352×288 352 × 288 416 × 240 416 ×240 832×480 832 × 480 1280×720 1280×720 1920×1080 1920 ×1080 2560×1600 2560×1600

Frame Rate 30fps 30 fps 50 fps 30 fps 50 fps 60 fps 60 fps 60 fps 50 fps 60 fps 30 fps 30 fps

B. Experimental Results Fig. 9 shows that compared with the original frame (column 1st), the encrypted frame (column 2nd) has difficulty distinguishing the original authenticity in terms of texture and outline. In addition, the decrypted frames (column 3rd) are identical to the original frames. It indicates the proposed scheme can significantly affect the subjective vision of the video content.

(27)

VI. P ERFORMANCE EVALUATION A. Experimental Setup The experimental works are conducted on a machine comprising an environment configuration, as depicted in Table II. The profile used by HM17.0 [23] is encoder randomaccess main, in which the QP is set to 24, the number of encoding frames (FramesToBeEncoded) is set to 100, and SideHideFlag is set to false. Additionally, AES initial

C. Slice Analysis Fig. 10 shows the slice analysis for the Foreman video sequence. The video loses the 11th to 14th slices, and the decrypted content of the 15th frame is the same as that of the original frame (18th frame). The first and second columns are the results of the Foreman video in the 1st frame (the corresponding frame is also the 1st frame) and the 15th frame (the corresponding frame is the 18th frame). Starting from the 1st frame as a normal decrypted frame, some slices are lost along the way, and the subsequent decrypted frames are still exactly the same as the original corresponding frames.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

9

Fig. 9: Experimental results for the Johnny and BasketballDrive video sequences. The first and second rows are the Johnny video sequence (#1 and # 30 frame), and the third and fourth rows are the BasketballDrive video sequence (#1 and # 30 frame) with the proposed scheme.

D. Subjective Vision Analysis The outcomes for the subjective vision analysis are presented in Fig. 11. The proposed method significantly impacts the subjective vision of the video content compared with the original frame. Additionally, the decrypted frame is identical to the original frame, confirming the correctness of the decryption. Furthermore, the proposed method causes the same level of distortion as the RSVE scheme does, indicating that it does not affect the subjective distortion levels of the video content. Fig. 10: Slice analysis: Foreman video original frame, encrypted frame and decrypted frame. E. Objective Indicator Analysis Furthermore, the PSNR and SSIM indicators for the Foreman video in Table IV show that the 15th decrypted frame and the original corresponding frame (18th frame) are the same. Thus, the random sequence generated by the proposed scheme is synchronized with the slice. TABLE IV: Slice analysis: PSNR and SSIM results Video Frame 1 (1) 15 (18)

Original 40.1872 37.7070

PSNR Encrypt 5.8797 6.6155

Decrypt 40.1872 37.7070

Original 0.9633 0.9477

SSIM Encrypt 0.2207 0.2138

Decrypt 0.9633 0.9477

The evaluation uses two indicators: the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). As shown in Table V, all the results are averaged over 100 frames for PSNR and SSIM. Notably, the proposed scheme shows significantly lower PSNR values and performs second in SSIM when compared to the state-of-the-art methods like Jovanovic et al. [26], Sallam et al. [18], and the RSVE scheme of Chan et al. [17]. This indicates that the proposed scheme can produce the best-distorted image, but the RSVE method provides better outcomes in terms of visual perception.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

10

Fig. 11: Experimental results for subjective vision analysis. The first and second rows are the KristenAndSara video sequence (#1 and # 30 frame), and the third and fourth rows are the Traffic video sequence (#1 and # 30 frame). The first to fourth columns are the original frame, the RSVE scheme, the proposed scheme, and the decrypted frame, respectively.

F. Edge Detection Analysis To avoid revealing important information from the edge of the image, the encrypted edge frame should have a strong distortion effect to resist edge detection attacks. It is evident from Fig. 12 that, compared with the original edge frame, the encrypted edge frame has a strong degree of distortion; hence, the content cannot be tracked; this shows the significance of the proposed scheme in effectively hiding edge information visually. Furthermore, Table VI displays the evaluation of the edge difference ratio (EDR). All the results are averaged over 100 frames of EDR. The EDR indicators for the encrypted edge frames are close to 1. In comparison, to the RSVE scheme, the proposed scheme shows slightly better performance due to the consideration of coefficient scrambling conditions. G. Chosen-Plaintext Attack Analysis The attacker creates some plaintext to generate the corresponding ciphertext and tries various means to obtain the correct key through ciphertexts. This type of attack is called a ”chosen plaintext attack”. HDavg is computed to calculate the difference between two sets of random sequences and to evaluate the resistance to chosen plaintext attacks, as shown in Table VII. The HDavg obtained for each group of videos is close to 0.5, indicating that there are large differences between the random sequences generated by each group of videos. The proposed scheme uses the SHA-384 hash value of the

Slice Header to update the key and initial vector, so each slice has a different random sequence, indicating that the proposed scheme can effectively resist chosen plaintext attacks. H. Replacement Attack Analysis Attackers may try to modify the encrypted content in an attempt to find some clues about the encrypted video, which is called a ”substitution attack”. The experiment involves four substitution attacks on the PartyScene video for testing, namely: • Replacing the encrypted Luma IPM with the Luma IPM as the first of the three most likely candidate prediction modes for the neighboring PU. • Replacing the encrypted QTC symbol with the QTC symbol =1 • Replacing the encrypted MVD symbol with the MVD symbol =1 • Replacing the encrypted MVD suffix with an MVD suffix =0. From Fig. 13, it can be observed that the results of trying to use four alternative attacks still cannot reveal some information related to the original frame. Table VIII portrays the results when the PSNR and SSIM indicators are used, considering that each frame is the result of the 30th frame of the PartyScene video. The decrypted frames of the four substitution attacks have smaller PSNR and SSIM values than the original frames do, and the PSNR and SSIM values are similar to each

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

11

TABLE V: PSNR and SSIM results for various encryption algorithms PSNR SSIM Jovanović Sallam RSVE Jovanović Sallam RSVE Proposed Original [26] [18] [17] [26] [18] [17] 12 51.2080 10.8199 10.1993 9.9067 9.3520 0.9984 0.0948 0.0741 0.0592 Mobile 24 40.9997 11.3320 11.0287 9.9786 9.9557 0.9889 0.0957 0.1092 0.0663 36 32.9772 10.2106 10.4236 9.5058 10.4717 0.9313 0.1098 0.1446 0.0738 12 51.5915 12.2185 13.4685 11.1852 10.5552 0.9961 0.3556 0.3765 0.2381 Foreman 24 44.4308 13.8540 11.9037 12.4377 7.9867 0.9712 0.3005 0.3830 0.2279 36 38.4626 13.0729 13.8619 12.7978 8.3192 0.9015 0.4150 0.4163 0.3437 12 51.0548 13.1237 13.5211 10.7650 11.1731 0.9977 0.1812 0.1887 0.1693 BlowingBubbles 24 41.5218 13.0176 13.9822 11.4384 10.9450 0.9701 0.1446 0.2549 0.1697 36 34.7572 12.2965 12.5875 10.7501 10.4957 0.8548 0.2278 0.2706 0.1648 12 51.3729 11.2945 11.9913 10.0016 10.8395 0.9975 0.2124 0.2068 0.1749 RaceHorses 24 42.0433 11.9281 11.0942 10.7121 11.1435 0.9815 0.2025 0.2412 0.1730 36 34.7851 12.0602 11.5157 10.8476 9.3066 0.8797 0.2498 0.2985 0.1797 12 51.0801 13.6093 13.5525 11.9879 10.7588 0.9984 0.1160 0.1481 0.1099 PartyScene 24 40.9438 13.3090 13.1844 11.9118 10.3023 0.9785 0.1339 0.1712 0.1054 36 33.7484 13.3780 13.1568 11.7889 10.3876 0.8724 0.1466 0.2068 0.1279 12 51.3380 11.4313 11.9658 11.2467 11.2885 0.9965 0.3335 0.3466 0.3023 BasketballDrillText 24 42.9425 12.0454 12.3082 9.9141 10.4729 0.9658 0.3936 0.4042 0.2728 36 36.3617 11.6326 11.9429 10.2188 11.5484 0.8780 0.4068 0.4683 0.3670 12 51.8393 14.0898 14.3128 13.2773 8.6507 0.9944 0.4925 0.5124 0.4359 Johnny 24 46.6292 14.2316 13.6275 12.9926 7.7676 0.9725 0.4706 0.4471 0.3682 36 41.9026 14.3882 14.9000 13.5148 9.1673 0.9428 0.5706 0.4749 0.3905 12 51.8475 15.7376 15.0158 14.0696 6.6476 0.9944 0.4474 0.4690 0.4022 KristenAndSara 24 46.4661 15.7631 15.3568 14.7831 6.8164 0.9765 0.4253 0.4882 0.3939 36 41.4027 15.1599 15.7244 13.9391 9.8735 0.9517 0.4779 0.4899 0.4221 12 51.0072 14.8807 14.4820 14.7322 10.9528 0.9951 0.5278 0.5424 0.3372 BasketballDrive 24 44.0864 13.2471 14.9034 14.5187 11.1040 0.9419 0.4941 0.5185 0.3726 36 40.1487 14.8385 14.6580 14.3899 11.3314 0.8975 0.5785 0.5943 0.4335 12 51.5661 13.1145 13.9256 12.4986 8.6248 0.9975 0.2110 0.2696 0.1761 BQTerrace 24 42.7011 15.5179 16.8617 14.2621 8.4283 0.9735 0.2548 0.2863 0.1904 36 37.0627 16.5752 16.7083 15.9872 8.6667 0.8937 0.3156 0.2922 0.2348 12 51.1566 12.5901 11.6006 11.1289 9.9424 0.9960 0.3295 0.3392 0.2888 Traffic 24 43.0685 11.5806 11.1236 11.0540 8.9662 0.9774 0.3464 0.3574 0.2725 36 37.4918 13.9138 13.6104 12.9100 9.3449 0.9193 0.3573 0.3852 0.3075 12 51.3575 13.2091 12.6736 12.4602 9.1224 0.9965 0.2911 0.3119 0.2442 PeopleOnStreet 24 44.2459 13.8099 13.3416 12.9737 9.2265 0.9774 0.3171 0.3422 0.2437 36 39.4948 13.7879 12.6562 11.8591 9.3852 0.9137 0.3591 0.3572 0.2943 PSNR: Peak Signal-to-Noise Ratio, QP: Quantization Parameter, RSVE: Robust Selective Video Encryption Video Sequence

QP

Original

TABLE VI: Edge difference ratio results Video Sequence Mobile Foreman BlowingBubbles RaceHorses PartyScene BasketballDrillText Johnny KristenAndSara BasketballDrive BQTerrace Traffic PeopleOnStreet

Original 0.0569 0.0853 0.1037 0.1091 0.0719 0.1050 0.1291 0.0818 0.1206 0.0856 0.1361 0.1437

Jovanović [26] 0.9219 0.8833 0.8846 0.9241 0.8923 0.8256 0.8380 0.8043 0.8980 0.8931 0.8896 0.9011

EDR Sallam [18] 0.9172 0.8993 0.8917 0.9234 0.8933 0.8347 0.8441 0.7995 0.9001 0.8969 0.8866 0.9003

RSVE [17] 0.9339 0.9345 0.9190 0.9378 0.9243 0.8885 0.9034 0.8711 0.9454 0.9171 0.9228 0.9311

Video Sequence 2

Resolution

Mobile BlowingBubbles PartyScene Johnny BasketballDrive Traffic

Foreman RaceHorses BasketballDrillText KristenAndSara BQTerrace PeopleOnStreet

352×288 416×240 832×480 1280×720 1920×1080 2560×1600

0.0819 0.0914 0.1095 0.2860 0.2514 0.2879 0.1935 0.1946 0.2085 0.2097 0.1991 0.2286 0.1380 0.1504 0.1498 0.3441 0.3514 0.4133 0.4259 0.4208 0.5031 0.3665 0.4072 0.4959 0.4328 0.4826 0.5282 0.2170 0.2510 0.2651 0.2919 0.3204 0.3286 0.2397 0.2455 0.2605

TABLE VIII: PSNR and SSIM results of PartyScene video under different attack conditions Proposed 0.9296 0.9271 0.9241 0.9415 0.9216 0.8921 0.9121 0.8807 0.9447 0.9146 0.9256 0.9349

TABLE VII: Average hamming distance results for all two different videos with the same resolution Video Sequence 1

Proposed

HDavg RSVE [17] Proposed 0.4968 0.4951 0.4940 0.4952 0.4926 0.4939 0.4991 0.4882 0.4981 0.4971 0.4931 0.4906

other compared with those of the encrypted frames. Thus, the proposed scheme can effectively resist substitution attacks.

Video Status Original Encrypt Replacement Attack 1 Replacement Attack 2 Replacement Attack 3 Replacement Attack 4

PSNR 34.1943 10.8279 11.2205 10.0635 10.7949 10.8971

SSIM 0.9457 0.1489 0.1497 0.2129 0.1464 0.1545

I. Information Entropy Analysis

Information entropy, also known as Shannon entropy, is used to measure the uncertainty of an event. The greater the information entropy, the more chaotic it is, and the probability of each event being average. Table IX shows the information entropy index analysis results for various videos of 100 frames. The results indicate that the proposed scheme maintains a significant level of randomness in pixel distribution. However, the RSVE scheme demonstrates superior performance in terms of information entropy indices.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

12

Fig. 12: Comparison of the edge images of KristenAndSara and the traffic sequence with the RSVE scheme. The first and second rows are the KristenAndSara video sequence (#1 and # 30 frame), and the third and fourth rows are the Traffic video sequence (#1 and # 30 frame). The first to third columns are the original frame, the RSVE scheme, and the proposed scheme frame, respectively.

Fig. 13: Replacement attack analysis of the PartyScene video sequence (# 30 frame). Original frame (upper left), encrypted frame (upper middle) and decrypted frames of the four substitution attacks.

J. Bit Rate Overhead Analysis TABLE IX: Information entropy results Video Sequence Mobile Foreman BlowingBubbles RaceHorses PartyScene BasketballDrillText Johnny KristenAndSara BasketballDrive BQTerrace Traffic PeopleOnStreet

Original 7.6477 7.2290 7.3596 6.9685 7.4359 6.9401 6.8321 7.1588 7.0192 7.2192 7.3522 7.0169

Information Entropy Jovanović [26] Sallam [18] 7.8328 7.8356 7.4967 7.5597 7.5867 7.4515 7.6297 7.6359 7.8441 7.8611 7.6065 7.8590 6.9636 6.8960 7.1632 7.1760 7.4958 7.3175 7.7355 7.7098 7.8077 7.8345 7.8669 7.8392

RSVE [17] 7.9159 7.6519 7.6711 7.7302 7.8621 7.5805 7.0704 7.4700 7.6533 7.8445 7.9027 7.8877

Proposed 7.6351 7.1786 7.6489 7.7367 7.8091 7.6999 6.9851 7.3915 7.3984 7.7709 7.7848 7.8615

The values for the bit rate overhead, an important parameter for effective video encryption, are shown in Table X. When coefficient scrambling is considered, the proposed scheme outperforms the RSVE scheme by an average of 45.13% in terms of bit rate overhead. This suggests that the bit rate overhead can be significantly improved compared with that of the RSVE scheme.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

TABLE X: Bit rate overhead results Video Sequence Mobile Foreman BlowingBubbles RaceHorses PartyScene BasketballDrillText Johnny KristenAndSara BasketballDrive BQTerrace Traffic PeopleOnStreet

Bit Rate Overhead RSVE [17] Proposed 0.0566 0.0187 0.0714 0.0426 0.0666 0.0350 0.0661 0.0286 0.0706 0.0352 0.0799 0.0481 0.0764 0.0582 0.0866 0.0590 0.0861 0.0424 0.0566 0.0323 0.0697 0.0416 0.0767 0.0380

Improvement Rate (%) 66.96 40.34 47.45 56.73 50.14 39.80 23.82 31.87 50.75 42.93 40.32 50.46

K. Key Space Analysis The attacker tries every possible key to crack the key, such attack is called a ”brute force attack”. To resist brute force attacks, the key space must be very large. The proposed scheme uses the AES encryption algorithm to create a random sequence. With a key size of 256 bits, there are 2256 possible combinations required to crack the key, indicating that the proposed scheme has an adequate ability to withstand attacks. L. Encryption Time Overhead Analysis The proposed algorithm demonstrates improved performance for low-resolution video sequences, making it suitable for mobile devices. Table XI indicates that the proposed scheme performs comparably to the RSVE scheme across 12 videos. However, when considering the average encryption time overhead for all the videos, the RSVE scheme has an overhead of approximately 17.05%, while the proposed scheme has only about 1.91%. This finding indicates that the proposed scheme is highly efficient, offering better performance in terms of encryption time overhead. TABLE XI: Encryption time overhead results Video Sequence Mobile Foreman BlowingBubbles RaceHorses PartyScene BasketballDrillText Johnny KristenAndSara BasketballDrive BQTerrace Traffic PeopleOnStreet

Overhead RSVE [17] Proposed 0.0207 0.008 0.0211 0.008 0.0127 0.010 0.0227 0.016 0.0174 0.009 0.0186 0.011 0.0043 0.016 0.0199 0.027 0.0227 0.024 0.0166 0.034 0.0065 0.036 0.0214 0.030

M. NPCR and UACI Analysis Table XII and Table XIII present a comparison of the NPCR (Number of Pixels Change Rate) and UACI (Unified Average Changing Intensity) results for various video sequences, alongside findings from state-of-the-art works, including Wallendael et al. [9], Sallam et al. [18], and Boyadjis et al. [16]. The proposed scheme demonstrates optimal performance, outperforming the other approaches in terms of NPCR. Additionally, the UACI results are commendable, as they closely align with the expected value 0.334635.

13

TABLE XII: Comparison NPCR results for various video sequences: proposed scheme v/s sota Video Sequence

Original

Mobile Foreman PartyScene Johnny BasketballDrive

0.72235 0.70214 0.72662 0.65342 0.69969

Wallendael [9] 0.99559 0.99569 0.99510 0.99419 0.99509

NPCR Sallam [18] 0.99562 0.99565 0.99515 0.99384 0.99520

Boyadjis [16] 0.99569 0.99592 0.99498 0.99386 0.99417

Proposed 0.99437 0.99837 0.99665 0.99801 0.99612

TABLE XIII: Comparison UACI results for various video sequences: proposed scheme v/s sota Video Sequence

Original

Mobile Foreman PartyScene Johnny BasketballDrive

0.01533 0.00982 0.01509 0.00574 0.00814

Wallendael [9] 0.28636 0.29351 0.24876 0.26602 0.23606

UACI Sallam [18] 0.28667 0.29340 0.24931 0.26903 0.23898

Boyadjis [16] 0.29201 0.29130 0.24495 0.26239 0.21722

Proposed 0.26266 0.452675 0.32645 0.397382 0.264240

N. Comparison 1) Computational Complexity: Table XIV analyzes the complexities of the proposed approach compared to stateof-the-art models. It considers N as the number of syntax elements or transform coefficients processed, M as the number of semantic features or elements extracted, P as the parameters related to steganography payload size or hashing complexity, K as the size parameter for SHA-256, and H as the hierarchical levels in encryption. The computational complexities increase in the following order: Proposed Scheme ≡ [27] ≡ [10] < [17] < [28] < [29] < [30]. TABLE XIV: Comparison: computational complexity Contributor Sheng et al. [27] Chen et al. [17] Sheng et al. [29] Peng et al. [10] Sheng et al. [30] Xing et al. [28] Proposed Scheme

Computational Complexity O(N ) O(N ) + O(K) O(N ) + O(M ) O(N ) O(N ) + O(P ) + O(N log N ) O(N ) + O(H) O(N )

2) Comparative Analysis: The proposed scheme is evaluated and compared with other related H.265 SEA studies using six key criteria (C1–C6) to ensure both security and practicality. C1 (Format Compliance) checks whether the encrypted bitstream still adheres to the H.265/HEVC standard, √ allowing playback on standard decoders. It is marked as (Yes) if format compliance is preserved, or × (No) if the encryption breaks standard compliance. C2 (Distortion Level of the Encrypted Video) measures how unintelligible the video becomes without decryption. It is marked as ↑ (High) for strong distortion with excellent confidentiality, ⇝ (Medium) for partial distortion with moderate confidentiality, and ↓ (Low) for weak distortion with poor confidentiality. A higher distortion level is desirable, as it means the encrypted video is more secure and unintelligible to unauthorized viewers. If the distortion is low, the content may still be recognizable, leading to poor confidentiality. C3 (Execution Time Overhead) assesses the extra computational cost introduced by encryption. It is marked as ↑ (High) for heavy time overhead (inefficient), ⇝ (Medium) for moderate overhead (acceptable), and ↓ (Low) for minimal

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

14

TABLE XV: Comparative analysis: proposed scheme v/s state-of-the-art performance parameters Contributor Lee et al. [2] Zhang et al. [12] Zhang et al. [19] Sallam et al. [11] Wen et al. [13] Liu et al. [21]

Ye et al. [4] Peng et al. [10]

Chen et al. [17] Sheng et al. [29] Xing et al. [28]

Encryption Method NALU Header MVD, QTC, Delta QP MVD, QTC, Delta QP QTC, MVD, Delta QP, SAO sign, Reference frame index, Residual size QTC sign, Luma IPM QTC, MVD, Delta QP sign, SAO sign, Merge index, Reference frame index, Partition mode MVD, QTC, Delta QP Luma IPM, Chroma IPM, QTC, MVD, Merge index, MVP index, Reference frame index, SAO, Coefficient scrambling QTC, MVD, Luma IPM, Coefficient scrambling Chaos-Based Tunable Selective Encryption with Semantic Understanding Hierarchical Multiscenario HEVC Video Encryption Scheme

Sheng et al. [30]

Deep Hashing Network and Steganography Based Selective Encryption

Sheng et al. [27]

Content-Aware Tunable Selective Encryption using Sine-Modular Chaotification Model QTC, MVD, Luma IPM, Coefficient scrambling

Proposed Scheme

Cipher Method AES-ECB AES-CTR Lorenz chaotic map RC6

Year 2020 2020 2021 2018

AES-CTR Integer dynamic tent mapping

2022 2020

L-ICMIC-CML chaotic map AES-CTR

2021 2020

Improved RC4

2023

Logistic map-based chaos, Semantic content analysis Hierarchical selective encryption with chaotic sequences and symmetric cipher Deep hashing, Steganographic embedding, Symmetric cipher Chaotic map (Sine-Modular) based coefficient scrambling

2024

AES-CTR

2025

↑: High, ↓: Low, ⇝: Moderate,

overhead (best performance). Lower overhead is preferable for real-time or large-scale applications, such as streaming, while high overhead can make the system impractical in real-world deployments. C4 (Bit Rate Overhead) considers the impact on compression efficiency, as a minimal increase in bit rate preserves storage and transmission efficiency. It is marked as ↑ (High) for a significant increase in bit rate (poor efficiency), ⇝ (Medium) for a moderate increase (tolerable), and ↓ (Low) for minimal increase (best case). Minimizing bit rate overhead ensures that encrypted videos remain efficient to store and transmit, whereas high overhead may strain bandwidth and storage. C5 (Robustness to Slice Loss) evaluates whether decryption remains functional even if part of the video stream is lost, √ ensuring reliability in practical networks. It is marked as (Yes) if decryption remains functional despite slice loss, and × (No) if decryption fails or produces unusable results after slice loss. Finally, C6 (Resistance to Chosen-Plaintext Attack) determines the scheme’s resilience against adversaries attempting to exploit known√input-output pairs to break the encryption. It is marked as (Yes) if it is resistant to CPA (high security) and × (No) if it is vulnerable to CPA (weak security). Together, these six metrics comprehensively balance compatibility, confidentiality, efficiency, robustness, and security. The comparison results regarding format compatibility, visual distortion, chosen-plaintext attack, bit rate overhead, and robustness are presented in Table XV. Most SEAs have characteristics that satisfy C1 and C3, but most of them are still insufficient in C2. Additionally, some SEAs that meet C2 still do not perform well in C4. The coefficient scrambling scheme in the RSVE method leads to a significant bit rate overhead. Therefore, the scheme proposed in this paper aims to satisfy C2 and all other five performance criteria. Consequently, the proposed scheme is competent for other H.265 SEA research. This reflects the significant reliability of the proposed approach

√

2024

2024

2025

C1 × √ √ √ √ √

√ √

√ √ √

× √

√

C2 − ↓ ↓ ⇝

C3 ↓ ↓ ⇝ ↓

C4 ↓ ↓ ↓ ↓

C5 × × × ×

C6 × × × ×

⇝ ↓

↓ ↓

↓ ↓

× ×

× ×

↓ ↑

↓ ↑

↓ ↑

× ×

× ×

↑

↓

⇝

√

√

↑

⇝

⇝

↑

⇝

⇝

↑

↑

↑

×

↑

⇝

↓

×

↑

↓

↓

× √

√

√ √

√

√

√

: Yes, ×: No, −: Unknown

for practical applications. VII. D ISCUSSION AND L IMITATIONS The evaluation of the proposed chaotic map-based video encryption scheme across multiple quality, security, and efficiency metrics reveals that, while it demonstrates strong performance in several areas, it does not consistently outperform existing approaches in all scenarios. Each metric highlights specific strengths and trade-offs of the design. The higher PSNR after decryption indicates that the proposed scheme preserves visual quality more effectively than competing methods. This advantage arises from the selective and chaosdriven encryption design, which perturbs critical coefficients without introducing excessive reconstruction noise. As a result, the decrypted video closely matches the original content. However, despite the strong PSNR, the scheme underperforms in SSIM, suggesting that while overall pixel-level fidelity is high, subtle structural information and texture consistency are not fully preserved. This discrepancy may be attributed to chaotic modifications of coefficients, which can distort local structures and edges in ways that PSNR does not capture. SSIM’s sensitivity to structural degradation explains why it presents lower scores. Additionally, the encrypted frames exhibit strong edge preservation, indicating that the chaotic scrambling does not oversmooth or blur significant features. This is particularly beneficial for applications where edge integrity is critical, such as surveillance. However, this may also explain the weaker SSIM, as excessive edge contrast can lead to reduced perceived structural similarity. Moreover, the scheme effectively resists chosen-plaintext attacks, owing to the sensitivity of chaotic maps to initial conditions and keys. Even minor differences in input lead to vastly different encrypted outputs, reinforcing its cryptographic strength. This characteristic shows that the design is well-suited against

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

advanced adversarial models. However, the entropy values of the encrypted frames are lower, suggesting incomplete randomness in the ciphertext distribution. This limitation arises because the chaotic map, while non-linear, may introduce statistical bias in certain regions of the frame. The partial determinism in coefficient modification could reduce randomness, allowing for the presence of residual patterns. The scheme introduces minimal bit rate expansion compared to RSVE and similar methods. By carefully preserving compression-friendly structures in the H.265 stream, it avoids disrupting entropy coding efficiency, ensuring better bandwidth utilization for practical deployment. The computational burden of chaotic mapping is comparable to that of the RSVE scheme, balancing complexity with efficiency. While it does not outperform RSVE, it shows that the proposed scheme maintains realtime feasibility. This parity results from iterative chaotic computations, which, although secure, do not provide a speed advantage over lighter alternatives. Additionally, a high NPCR indicates strong sensitivity to plaintext changes, meaning the encryption effectively diffuses small input variations across the entire frame. The chaotic diffusion mechanism enhances pixel-level unpredictability, contributing to robustness against differential attacks. Although the NPCR is strong, the UACI results are comparable to existing methods, suggesting that while the encryption process introduces frequent pixel value changes, the magnitude of these changes is not sufficiently large across the image domain. This indicates that the diffusion effect is strong in terms of pixel position changes but limited in terms of altering pixel intensity amplitude. This can be explained by the selective encryption approach, in which only certain coefficients are modified, leading to limited amplitude variation. Overall, the inconsistency in performance across metrics reflects a core trade-off: the scheme prioritizes preserving compression efficiency (with low bit rate overhead) and ensuring strong cryptographic resistance (NPCR) at the expense of structural similarity (SSIM), entropy randomness, and UACI. Specifically, reliance on chaotic maps while beneficial for diffusion and attack resistance may create predictable statistical behavior in certain coefficient domains, reducing overall entropy. Similarly, selective encryption preserves format and bandwidth efficiency but inherently sacrifices structural integrity and complete intensity diffusion. The mixed performance of the proposed scheme highlights the need for targeted improvements to achieve more consistent results across all metrics. To address the underperformance in SSIM, information entropy, and UACI, future work could explore hybrid chaotic diffusion models that combine pixel and transform domain operations to better preserve structural similarity while enhancing randomness. Incorporating adaptive coefficient selection may balance distortion strength with compression efficiency, improving both SSIM and entropy. Furthermore, integrating multi-round chaotic transformations or entropy-boosting post-processing steps could enhance ciphertext unpredictability without significantly increasing bit rate or execution time. These refinements would strengthen the scheme’s robustness while maintaining its current advantages in PSNR, edge preservation, NPCR, and bit rate efficiency.

15

VIII. C ONCLUSION AND F UTURE W ORK This paper presents a coefficient scrambling scheme based on a chaotic map by devising scrambling conditions for the coefficients wisely. It specifically focuses on encrypting three types of information: motion vector differences (MVD), quantization table coefficients (QTC), and Luma IPM (Luma Intra Prediction Modes). Additionally, the scheme synchronizes the generated random sequence with the slice and links it to the H.265 encoded stream. The experimental results demonstrate that the proposed scheme offers high security, compliance with formats, fast execution time, synchronous update with the slice, and resistance to common attacks. Compared with the RSVE scheme, it reduces the average bit rate overhead. Therefore, the proposed coefficient scrambling scheme presents a superior option for enhancing the security of H.265/HEVC video SEA. Future work will focus on developing a video encryption algorithm for critical areas via a 2D extended Schaffer function map and neural networks. Additionally, temporal action segmentation will be integrated to improve the accuracy, robustness, and adaptability of the model for video streams. R EFERENCES [1] G. J. Sullivan, J.-R. Ohm, W.-J. Han, and T. Wiegand, “Overview of the high efficiency video coding HEVC standard,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 22, no. 12, pp. 1649– 1668, 2012. [2] M. K. Lee and E. S. Jang, “Start code-based encryption and decryption framework for HEVC,” IEEE Access, vol. 8, pp. 202 910–202 918, 2020. [3] M. Farajallah, W. Hamidouche, O. Déforges, and S. E. Assad, “ROI encryption for the HEVC coded video contents,” in 2015 IEEE International Conference on Image Processing (ICIP), 2015, pp. 3096–3100. [4] Q. Ye, Q. Zhang, S. Liu, and K. Chen, “A novel chaotic system based on coupled map lattice and its application in HEVC encryption,” Mathematical Biosciences and Engineering, vol. 18, no. 6, pp. 9410– 9429, 2021. [5] S. Gao, Z. Zhang, Q. Li, S. Ding, H. H.-C. Iu, Y. Cao, X. Xu, C. Wang, and J. Mou, “Encrypt a story: A video segment encryption method based on the discrete sinusoidal memristive rulkov neuron,” IEEE Transactions on Dependable and Secure Computing, pp. 1–15, 2025. [6] S. Gao, R. Wu, H. H.-C. Iu, U. Erkan, Y. Cao, Q. Li, A. Toktas, and J. Mou, “Chaos-based video encryption techniques: A review,” Computer Science Review, vol. 58, p. 100816, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1574013725000929 [7] X. Zhu, F. Liu, and D. Hu, “H.265/HEVC: New video coding standard and its extension,” 2016. [8] M. K. Lee and E. S. Jang, “Cryptanalysis of start code-based encryption method for HEVC,” IEEE Access, vol. 9, pp. 92 568–92 577, 2021. [9] G. Van Wallendael, A. Boho, J. De Cock, A. Munteanu, and R. Van De Walle, “Encryption for high efficiency video coding with video adaptation capabilities,” IEEE Transactions on Consumer Electronics, vol. 59, no. 3, pp. 634–642, 2013. [10] F. Peng, X. Zhang, Z.-X. Lin, and M. Long, “A tunable selective encryption scheme for H.265/HEVC based on Chroma IPM and Coefficient Scrambling,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 30, no. 8, pp. 2765–2780, 2020. [11] A. I. Sallam, E.-S. M. El-Rabaie, and O. S. Faragallah, “CABAC-based selective encryption for HEVC using RC6 in different operation modes,” Multimedia Tools and Applications, vol. 77, no. 21, pp. 28 395–28 416, 2018. [12] Q.-J. Zhang, Q. Ye, Z.-M. Yuan, and L. Li, “Fast HEVC selective encryption scheme based on improved CABAC coding algorithm,” in 2020 IEEE 6th International Conference on Computer and Communications (ICCC), 2020, pp. 1022–1028. [13] W. Wen, R. Tu, Y. Zhang, Y. Fang, and Y. Yang, “A multi-level approach with visual information for encrypted H.265/HEVC videos,” Multimedia Systems, vol. 29, no. 3, pp. 1073–1087, 2023.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 36, NO. 4, APRIL 2026

[14] M. Yang, L. Zhuo, J. Zhang, and X. Li, “An efficient format compliant video encryption scheme for HEVC bitstream,” in 2015 IEEE International Conference on Progress in Informatics and Computing (PIC), 2015, pp. 374–378. [15] D. Xu, “Commutative encryption and data hiding in HEVC video compression,” IEEE Access, vol. 7, pp. 66 028–66 041, 2019. [16] B. Boyadjis, C. Bergeron, B. Pesquet-Popescu, and F. Dufaux, “Extended selective encryption of H.264/AVC (CABAC)- and HEVCencoded video streams,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 27, no. 4, pp. 892–906, 2017. [17] C. Chen, X. Wang, G. Liu, and G. Huang, “A robust selective encryption scheme for H.265/HEVC video,” IEEE Access, vol. 11, pp. 17 252– 17 264, 2023. [18] A. I. Sallam, O. S. Faragallah, and E.-S. M. El-Rabaie, “Hevc selective encryption using rc6 block cipher technique,” IEEE Transactions on Multimedia, vol. 20, no. 7, pp. 1636–1644, 2018. [19] Q.-j. Zhang, Q. Ye, L. Li, S.-j. Liu, and K.-q. Chen, “An efficient selective encryption scheme for HEVC based on hyperchaotic lorenz system,” in 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC), 2021, pp. 683–690. [20] A. I. Sallam, E.-S. M. El-Rabaie, and O. S. Faragallah, “Efficient HEVC selective stream encryption using chaotic logistic map,” Multimedia Systems, vol. 24, no. 4, pp. 419–437, 2018. [21] B. Liu, J. Liu, S. Wang, M. Zhong, B. Li, and Y. Liu, “HEVC video encryption algorithm based on integer dynamic coupling tent mapping,” Journal of Advanced Computational Intelligence and Intelligent Informatics, vol. 24, no. 3, pp. 335–345, 2020. [22] Q. Lu, C. Zhu, and X. Deng, “An efficient image encryption scheme based on the LSS Chaotic Map and Single S-Box,” IEEE Access, vol. 8, pp. 25 664–25 678, 2020. [23] “HM 17.0 Github.” [Online]. Available: https://vcgit.hhi.fraunhofer.de/ jvet/HM/-/tree/HM-17.0 [24] “YUV Video Sequences.” [Online]. Available: https://media.xiph.org/ video/derf/ [25] “HEVC Test Sequences.” [Online]. Available: http://ftp.kw.bbc.co.uk/ hevc/hm-10.0-anchors/bitstreams/ [26] B. Jovanović and S. Gajin, “An efficient mechanism of cryptographic synchronization within selectively encrypted h.265/hevc video stream,” Multimedia Tools and Applications, vol. 77, no. 2, pp. 1537–1553, 2018. [27] Q. Sheng, C. Fu, Z. Lin, J. Chen, X. Wang, and C.-W. Sham, “Contentaware tunable selective encryption for hevc using sine-modular chaotification model,” IEEE Transactions on Multimedia, vol. 27, pp. 41–55, 2025. [28] M. Xing, H. Yu, W. Zhang, and Z. Zhu, “A hierarchical multiscenario h.265/hevc video encryption scheme,” International Journal of Bifurcation and Chaos, vol. 34, no. 01, p. 2450013, 2024. [Online]. Available: https://doi.org/10.1142/S0218127424500135 [29] Q. Sheng, C. Fu, M. Tie, X. Wang, J. Chen, and C.-W. Sham, “A chaos-based tunable selective encryption algorithm for h.265/hevc with semantic understanding,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 11, pp. 11 040–11 055, 2024. [30] Q. Sheng, C. Fu, Z. Lin, J. Chen, X. Wang, and C.-W. Sham, “Content-aware selective encryption for h.265/hevc using deep hashing network and steganography,” ACM Trans. Multimedia Comput. Commun. Appl., vol. 21, no. 1, Dec. 2024. [Online]. Available: https://doi.org/10.1145/3698400

Liang-Wei Li has worked for his post graduation at Department of Computer Science and Engineering, National Sun Yat-sen University (NSYSU), Kaohsiung, Taiwan. His master’s thesis research was based on the video encryption, especially the encryption of H.265 encoding. He had excellent academic performance in college received three certificates of excellence, and ranked in the top three overall. Currently, he is working as a Firmware Engineer.

16

Chung-Nan Lee (Member, IEEE) received the B.S. and the M.S. degrees in electrical engineering from the National Cheng Kung University, Tainan, Taiwan, in 1980 and 1982, respectively, and the Ph.D. degree in electrical engineering from the University of Washington, Seattle, WA, USA, in 1992. Since 1992, he has been with the National Sun YatSen University, Kaohsiung, Taiwan. He served as Chairman of the Department of Computer Science and Engineering, from 1999 to 2001. Dr. Lee was the President of the Taiwan Association of Cloud Computing, from 2015 to 2017, and the VP for TA of Asia-Pacific Signal and Information Processing Association, from 2019 to 2020. In 2016, he received an outstanding engineering Professor Award from the Chinese Institute of Engineers, Taiwan. His research interests include multimedia over wireless networks, cloud computing, and IoT.

Kishu Gupta (Member, IEEE) received the Ph.D. degree in computer science from India in 2023. She is associated as a Post-Doctoral Researcher with National Sun Yat-sen University (NSYSU), Kaohsiung, Taiwan. She receiving a prestigious INSPIRE Fellowship sponsored by the DST, India. She has research findings published with top-notch venues, including IEEE TNNLS, IEEE TSMC, IEEE TASE, IEEE TITS, IEEE J-BHI, IEEE TCE, E-JNCA, Applied Soft Computing, Scientific Reports, Cluster Computing, and Procedia-Computer Science. Her major research interests include data security and privacy, evolutionary optimization, cloud computing, traffic management, federated learning, machine learning, neural networks, and quantum ML. Also, she was honored with a Gold Medal for her Academic Excellence (Ist Rank in M.Sc.), shortlisted for the Taiwan Comprehensive University System-Young Scholar (TCUS-YS) award, and received a Best Paper Award in RTIP2R-2024 Conference.

Huei-Fang Yang (Member, IEEE) received her Ph.D. in computer science from Texas A&M University, College Station, TX, USA, in 2011. She is an associate professor in the Department of Computer Science and Engineering at National Sun Yat-sen University, Taiwan. Her research interests include computer vision and deep learning. In recent years, her work has focused on image retrieval and medical image analysis.

Ashutosh Kumar Singh (Senior Member, IEEE) is working as a Professor and Director of Indian Institute of Information Technology Bhopal, India. Also, he is working as Adjunct Professor in the VIZJA University, Warsaw, Poland. He received his Ph.D. in Electronics Engineering from Indian Institute of Technology, BHU, India and Post Doc from Department of Computer Science, University of Bristol, UK. He has research and teaching experience in various Universities of the India, UK, and Malaysia. His research area includes Design and Testing of Digital Circuits, Data Science, Cloud Computing, Machine Learning, Security. He has published more than 400 research papers in different journals and conferences of high repute. His research paper, published in the IEEE Transactions on Cloud Computing Journal, was honored with the 2022 Best Paper Award by the IEEE Computer Society Publications Board.

Related documents

• Revisiting Code Debloating with Ground Truth-based Evaluation#120588
• The Base45 Data Encoding#502654
• Thread-Efficient Decoding for Neural Texture Compression#781170
• Thread-Efficient Decoding for Neural Texture Compression#779795
• Test Vectors for the Stream Cipher RC4#501394