Source-linked AI summary
Encryption-then-Compression Systems using Grayscale-based Image Encryption for JPEG Images
Tatsuya Chuman, Warit Sirichotedumrong, Hitoshi Kiya
TL;DR
The paper addresses secure image transmission through untrusted providers when JPEG-compatible Encryption-then-Compression is required. It proposes grayscale-based block scrambling with smaller blocks and less color information, and reports stronger resistance to ciphertext-only attacks while maintaining effective image quality after social-media processing.
Problem
Many EtC approaches rely on proprietary compression, while JPEG-based systems face block-size limitations caused by social-media recompression and color subsampling.
Method
The proposed scheme combines color channels into a grayscale-based image and applies block scrambling, rotation, inversion, and negative-positive transformation using Bx = By = 8.
Results
The proposed scheme enhanced security against ciphertext-only attacks and produced higher decrypted-image PSNR than conventional processing in Facebook experiments, with similar tendencies on Twitter.
Takeaways & Limitations
Smaller blocks, more encrypted blocks, and less color information made encrypted images more difficult to assemble with jigsaw puzzle solvers, including when Bx = By = 16.
Abstract
from arXiv · showhide
A block scrambling-based encryption scheme is presented to enhance the security of Encryption-then-Compression (EtC) systems with JPEG compression, which allow us to securely transmit images through an untrusted channel provider, such as social network service providers. The proposed scheme enables the use of a smaller block size and a larger number of blocks than the conventional scheme. Images encrypted using the proposed scheme include less color information due to the use of grayscale images even when the original image has three color channels. These features enhance security against various attacks such as jigsaw puzzle solver and brute-force attacks. In an experiment, the security against jigsaw puzzle solver attacks is evaluated. Encrypted images were uploaded to and then downloaded from Facebook and Twitter, and the results demonstrated that the proposed scheme is effective for EtC systems.
I. INTRODUCTION
The paper motivates JPEG-compatible Encryption-then-Compression systems for secure image sharing through untrusted social-media providers. It proposes grayscale-based block scrambling to relax conventional block-size limits while improving security and preserving image quality.
- Motivation: EtC systems protect images before JPEG compression, but many prior approaches use proprietary compression incompatible with social-media services.JPEG compatibility is important for transmission through platforms such as social media.
- Security Enhancement: Using smaller blocks and more blocks, together with less color information, is intended to improve invisibility and resistance to jigsaw-puzzle and brute-force attacks.The proposed design relaxes the JPEG-related block-size limitation while enhancing security.
- Evaluation: The evaluation reports improved security against ciphertext-only attacks and effectiveness in image quality for EtC systems.The paper evaluates security and compression, including robustness to jigsaw puzzle solver attacks.
- Limitations of Conventional Encryption: Conventional block scrambling is limited to Bx = By = 16 because smaller blocks can cause JPEG compression loss and block distortion.With 8 × 8 blocks, chroma subsampling can combine low-correlation regions and create interpolation artifacts.
B. Security against Ciphertext-only Attacks
The paper evaluates security against ciphertext-only attacks, including brute-force and jigsaw puzzle solver attacks, under adversaries such as untrusted providers and unauthorized users.
- Security is evaluated against brute-force and jigsaw puzzle solver attacks as ciphertext-only attacks.The evaluation uses the ratio of correctly positioned blocks and computational cost for jigsaw puzzle solver robustness.
1) Security:
The encryption scheme combines block permutation and transformations, with a key space that grows factorially with the number of blocks and exponentially across per-block operations.
- The number of possible block permutations is n!, where n is the number of image blocks.
- The encryption pipeline applies four block scrambling-based processing steps to the divided image.
- The full key space combines block permutation, rotation and inversion, negative-positive transformation, and color-component shuffling.The combined expression is represented as NA(n).
- The conventional examples use 140 blocks at 16 × 16 and 560 blocks at 8 × 8, while jigsaw solving treats encrypted blocks as puzzle pieces.
- NA(n) is expressed as n! · 8^n · 2^n · 6^n for the complete encryption process.
3) Jigsaw Puzzle Solver Attack:
Jigsaw puzzle attacks become harder with many small blocks, JPEG distortion, and reduced color information; the proposed scheme targets these conditions through grayscale-based encryption.
- Jigsaw puzzle assembly is difficult when encrypted images contain many blocks, small blocks, and JPEG distortion.
- Reducing the number of color channels makes encrypted-image assembly more difficult because many solvers use color information.
- The proposed scheme has higher security than the conventional scheme because it provides a larger number of blocks and smaller block size.
- Known-plaintext robustness is addressed by assigning a different key to each image, while undisclosed keys avoid the considered chosen-plaintext attack.
C. Summary of Image Encryption for EtC Systems
The proposed EtC encryption uses grayscale-based processing to support smaller blocks under JPEG while reducing color information and improving security and image quality in social-media settings.
- Summary: The proposed properties enhance invisibility and security, improve robustness, and avoid the effect of JPEG color sub-sampling.
- Security: The proposed scheme uses 8 × 8 blocks instead of the conventional 16 × 16 minimum, enhancing robustness against ciphertext-only attacks.
- Procedure: An RGB image is split into channels, transformed to YCbCr, combined into one grayscale-based image, and then encrypted with block transformations.The grayscale-based image contains 3 × X × Y pixels, and the channel-combination arrangement has some freedom.
- JPEG compression: The proposed scheme uses JPEG lossy compression because social-network and cloud-photo services commonly use it.The encryption schemes are also applicable to lossless compression methods, which generate no distortion.
- JPEG compression: Grayscale-based encrypted images use one JPEG quantization table, whose selection affects compression performance.
C. Decompression and Decryption of Encrypted Images
The proposed grayscale-based encryption supports JPEG decompression without chroma interpolation and then reconstructs the color image through a six-step decryption process.
- JPEG decompression omits color interpolation for the proposed encrypted images because they contain no chroma component.
- Decryption divides the image into blocks, reverses block transformations and scrambling, separates grayscale data into Y, Cb, and Cr, converts to RGB, and combines the channels.The process uses secret-key components K1, K2, and K3.
- The larger block count and one-channel encrypted representation enhance resistance to brute-force and jigsaw puzzle solver attacks.The one-channel representation removes the three-channel information commonly used by puzzle solvers.
- The proposed scheme uses grayscale encrypted images, enabling 8 × 8 encryption blocks even with JPEG compression.
- Using 8 × 8 blocks increases the number of blocks fourfold relative to 16 × 16 blocks, while the grayscale-based construction yields 12 times as many blocks as the conventional scheme.
2) Large Number of Blocks :
The scheme expands the number of scrambled blocks and reduces available color information, making ciphertext-only assembly and brute-force search more difficult while supporting privacy-preserving social-media transmission.
- The running time required by jigsaw puzzle solvers increases with the number of pieces, so more blocks make encrypted-image assembly more difficult.
- One-channel grayscale ciphertext removes the three color channels that most jigsaw puzzle solvers use to assemble blocks.
- The proposed key space NB(n) is larger because the scheme increases the number of blocks, strengthening robustness against brute-force attacks.
- EtC transmission keeps the secret key with the user, so social-media providers can decompress images without receiving the key.
- The evaluation considers uploaded and downloaded JPEG relationships involving sub-sampling ratios, quality factors, and provider resolution limits.
2) Recompression:
The proposed encrypted images preserve JPEG compression and decrypted quality more effectively than schemes that ignore JPEG processing, including after social-media recompression.
- The proposed scheme performs better in image quality because encrypted images avoid color sub-sampling during social-media processing.
- The experiments compress and decompress encrypted images, evaluate PSNR on 1,338 UCID color images, and compare rate-distortion behavior.
- As bitrate increases, the proposed encrypted images show better compression performance than non-encrypted images using 4:2:0 sub-sampling.
- Encryption methods that do not account for JPEG compression produce heavily degraded decrypted images, whereas the proposed scheme preserves nearly non-encrypted quality.
- Facebook and Twitter uploads and downloads were compared across proposed, conventional, and non-encrypted images.
1) Experimental Conditions:
The experiments measure compression and social-media robustness by encrypting, recompressing, downloading, decrypting, and comparing reconstructed images using PSNR.
- The evaluation uses the same dataset and JPEG compression parameters as the preceding compression-performance experiment.
- The procedure generates ciphertext, compresses it with Qfu, uploads and downloads it, decompresses and decrypts it, then computes PSNR against the original image.
- The study focuses on Facebook and Twitter because they recompress uploaded images under specified quality-factor conditions.
- For Facebook, the proposed scheme avoids interpolation and produces higher decrypted-image PSNR than the conventional scheme.
- For Twitter, images were uploaded at Qfu = 90, 95, and 100 because recompression depends on conditions including Qfu ≧85 with 4:2:0 sub-sampling.
- Twitter’s 4:2:0 downloads require user-side interpolation, resulting in lower decrypted-image PSNR than the proposed scheme.
- Most proposed-scheme PSNR values are higher with the luminance table than with the chrominance table because providers commonly recompress grayscale uploads using the luminance table.
C. Robustness against Jigsaw Puzzle Solver Attacks
The experiment evaluates how well an extended jigsaw puzzle solver can reconstruct encrypted images using direct, neighbor, and largest-component compatibility measures. It averages these measures over independently encrypted images generated from 20 resized UCID images.
- Dc measures the ratio of pieces placed in their correct positions.Dc(Id) averages the indicator that each piece occupies its correct location.
- Nc measures the ratio of correctly joined neighboring blocks.It evaluates whether boundaries between adjacent pieces are correctly formed.
- Lc measures the largest correctly connected component relative to the total number of image blocks.It uses the largest partial area with correct adjacencies.
- Higher Dc, Nc, and Lc values indicate greater compatibility between the assembled image and the encrypted image structure.All three measures range from 0 to 1.
- The solver selects the reconstruction with the highest combined Dc, Nc, and Lc score for each encrypted image.Forty encrypted images were generated per ordinary image using different keys, and the procedure was repeated independently.
2) Experimental results:
The proposed scheme makes encrypted images substantially harder to reconstruct with a jigsaw solver by combining smaller blocks, more encrypted blocks, and reduced color information. Experiments also indicate lower assembled-image scores, increased solving time, and effective JPEG EtC operation across social platforms.
- Security evaluation: Reducing encrypted images to one color channel made puzzle assembly more difficult than with the conventional scheme.The one-channel images received lower scores than images assembled from conventional encrypted images.
- Security evaluation: The proposed scheme produced much lower jigsaw-solver scores than the conventional scheme, including Lc = 0.021 for the conventional scheme at Bx = By = 16.Even the proposed scheme with Bx = By = 16 remained substantially harder to assemble.
- Running time: 166.11 minutes were required to solve proposed encrypted images with Bx = By = 8, while their assembled images had Lc = 0.002.The smaller-block configuration took longer to assemble than the conventional scheme.
- Design implications: The proposed scheme supports Bx = By = 8 instead of the conventional scheme’s smallest Bx = By = 16, improving resistance to ciphertext-only attacks.The smaller block size creates more blocks, while grayscale images reduce color information.
- Practical applicability: The proposed scheme avoids interpolation-related block distortion and was experimentally shown applicable to Twitter and Facebook EtC systems.The grayscale-based design improves image quality after social-media processing.