[Paper Review] Bitstream-Based JPEG Image Encryption with File-Size Preserving

Hiroyuki Kobayashi, Hitoshi Kiya|arXiv (Cornell University)|Aug 17, 2018

Chaos-based Image/Signal Encryption10 references4 citations

TL;DR

This paper proposes a bitstream-based JPEG encryption scheme that preserves the original file size by selectively encrypting only the additional bits of Huffman-coded coefficients while avoiding changes to marker codes through careful byte-stuffing handling. The method ensures exact file-size equivalence between original and encrypted images, enabling seamless integration in systems relying on fixed-size image transmission, such as cloud storage or network hooking mechanisms.

ABSTRACT

An encryption scheme of JPEG images in the bitstream domain is proposed. The proposed scheme preserves the JPEG format even after encrypting the images, and the file size of encrypted images is the exact same as that of the original JPEG images. Several methods for encrypting JPEG images in the bitstream domain have been proposed. However, since some marker codes are generated or lost in the encryption process, the file size of JPEG bitstreams is generally changed due to the encryption operations. The proposed method inputs JPEG bitstreams and selectively encrypts the additional bit components of the Huffman code in the bitstreams. This feature allows us to have encrypted images with the same data size as that recoded in the image transmission process, when JPEG images are replaced with the encrypted ones by the hooking, so that the image transmission are successfully carried out after the hooking.

Motivation & Objective

• To address the problem of file-size variation in existing bitstream-based JPEG encryption methods due to marker code generation or loss.
• To enable secure encryption of JPEG images in the bitstream domain without altering the file size, ensuring compatibility with existing transmission and storage systems.
• To develop a mechanism that prevents unintended marker code occurrences or omissions during encryption by analyzing byte-stuffing behavior in entropy-coded data.
• To allow secure image transmission via hooking mechanisms where encrypted images must retain the same size as originals.
• To achieve high encryption coverage on additional bits while preserving the integrity of JPEG format and decoding compatibility.

Proposed method

• The method processes JPEG bitstreams and identifies bytes that are part of Huffman codes or additional bits, analyzing their bit patterns to determine encryption feasibility.
• It excludes encryption of bytes that are fully Huffman code (all '1's), fully additional bits, '00' bytes following 'FF', or bytes where the Huffman code is all '1's, as these could trigger or disrupt byte stuffing.
• Encryption is applied only to additional bits in bytes where the Huffman code is not all '1's and the byte is not 'FF' followed by '00', thus avoiding marker code conflicts.
• The approach leverages the structure of JPEG entropy-coded data and the rules of byte stuffing to ensure that no new marker codes are accidentally generated during encryption.
• By restricting encryption to specific, safe regions of the bitstream, the method maintains the exact same file size as the original JPEG image.
• The scheme is designed to be compatible with standard JPEG decoders and does not require changes to the compression or decompression pipeline.

Experimental results

Research questions

• RQ1 Can JPEG image encryption in the bitstream domain preserve the original file size exactly, even after encryption?
• RQ2 How can encryption be performed without introducing or removing JPEG marker codes due to byte-stuffing effects?
• RQ3 What specific parts of the JPEG bitstream can be safely encrypted without altering the file size or breaking the format?
• RQ4 To what extent can additional bits in Huffman-coded coefficients be encrypted while maintaining file-size consistency?
• RQ5 Can this method be used effectively in systems requiring fixed-size image transmission, such as cloud storage or network hooking?

Key findings

• The proposed method achieves exact file-size preservation: encrypted JPEG images have identical file sizes to their original counterparts across all tested Q-factors (50, 80, 95).
• For the Lena image at Q=50, 98.6% of the additional bits in both DC and AC components were successfully encrypted, with only 197 bytes excluded from encryption.
• At Q=95, 93.8% of the additional bits were encrypted in the DC-only case, and 97.2% in the both-components case, showing high encryption coverage even at high quantization levels.
• Compared to prior methods, such as Cheng [17] and EtC [1], the proposed method achieved zero file-size difference (0-byte change), while other methods introduced changes of +2 to +1714 bytes.
• The method successfully prevents unintended marker code generation by avoiding encryption in critical byte-stuffing boundary conditions, ensuring format compatibility.
• The encrypted images remain decodable by standard JPEG decoders, and visual quality degradation is consistent with expected security levels, with minimal visible information leakage when only DC components are encrypted.