在全球数字化浪潮的推动下，数字图像作为信息的重要载体广泛应用于社交媒体、医疗、金融、军事等多个领域。数字图像在为人们带来传输和存储便利的同时，其安全性问题也日益凸显。面对安全挑战，我国颁布了《数据安全法》、《个人信息保护法》以及《中华人民共和国密码法》等。图像加密技术作为保护隐私的有效手段之一，通过特定的加密算法将数字图像可逆地转换为不可读或不可用的形式，以确保数字图像在存储与传输时得到全方位的保护。
JPEG 图像作为目前应用最广泛的图像格式之一，其隐私安全问题的讨论更具现实应用价值。相比于未压缩图像，具有严格编码格式的JPEG 图像加密不仅要考虑密文图像的安全性，还需要考虑文件大小和格式兼容性。除此之外，还需要考虑用户对密文图像可用性的需求（如快速识别和预览）。如何平衡JPEG 图像安全性和可用性的同时实现低文件扩展和格式兼容性，无疑是一项巨大的挑战。现有JPEG 图像加密算法在格式兼容性和文件扩展方面具有较好的性能，然而却保留了保留构造密钥的图像块特征值以及其他块特征值，让密文图像在选择明文攻击下被攻破。为了提高加密算法对JPEG 图像隐私保护能力，本文首先对保留块特征值的JPEG 加密算法进行安全性分析，随后设计三种不同的加密方案：一是改变密钥特征的加密算法，二是提高特征改变率的加密算法，三是在二的基础上迎合一些用户对密文可用性的需求，提出了JPEG 图像缩略图保持加密（JPEG thumbnail preserving encryption, JPEG-TPE）算法。具体的研究内容如下：
1. 块特征值保留的JPEG 图像加密算法安全性分析，并提出“块特征值改变率”指标。保留用于构造自适应加密密钥的块特征值让攻击者可以从密文图像中重构出相同密钥的明文图像并实施CPA 攻击；保留的其他块特征值有助于置乱序列的快速估计且可以在CPA 攻击后揭示原始图像轮廓。实验结果表明，自适应加密密钥可重构且块特征值保留的JPEG 图像加密算法在CPA 条件下易破解得到轮廓图像和（或）原始图像。更进一步，通过分析块置乱序列的估计正确率与块特征值改变率之间的线性关系，我们提出了针对块置乱与其他加密步骤结合的JPEG图像在自适应密钥可重构条件下抵抗CPA 攻击能力的指标–“块特征值改变率”。
2. 改变构造自适应密钥块特征的JPEG 图像加密算法。利用现有基于VLC 映射的可逆信息隐藏(Reversible data hiding,RDH) 方案，通过自嵌入RSV（run/size, value）改变构造自适应加密密钥的块特征值，在不增加密钥管理负担的同时提高密文图像抵抗CPA 攻击的能力。在分析现有基于VLC 映射RDH 方案中嵌入容量与文件大小之间的单调性后，计算得到文件大小减小的最大可嵌入RSV 数量阈值。在DC 和AC 加密后的图像中，随机自嵌入不多于阈值数量的RSV 并生成密文图像。实验表明，提出的算法在密文图像比原始图像减小1,000 比特的情况下，使自适应密钥直接从密文图像中正确重构的概率降低到接近于零。
3. 提高块特征值改变率的JPEG 图像加密算法。为了在不增加文件大小的同时提高块特征值改变率，首次提出了不分割RSV 的AC 系数 (AC coefficients group with unidivided RSV, ACG-URSV) 的概念，并设计了ACG-URSV 全局置乱方案。由于不同的ACG-URSV 中包含了不同数量、不同游程、不同非零AC 值，因此置乱相同长度的ACG-URSV 可以改变不同块特征值。另外，为了减小文件扩展，设计了基于DC 差值同类映射来旋转DC 系数的加密方案。实验结果表明，提出的算法的块特征值改变率至少增加了20%，文件大小平均减小了2.036%。
4. 提高块特征值改变率的JPEG 图像缩略图保持加密算法。针对该领域同样存在特征值改变率低和ACG-URSV 全局置乱加密中需要依赖固定频率而导致置乱元素减少的问题，设计了自适应ACG-URSV 全局置乱方案。为了保留原始图像的缩略图的同时产生较小的文件扩展，对DC 差分值而不是DC 系数值，进行和不变且考虑溢出的值重分配。实验结果表明，提出算法的缩略图质量基本与现有算法持平，但可置乱元素平均增加了1809 个，图像的文件扩展减小了14.77%，块特征值改变率提高了38.28%，且不随加密图像块大小的增加而降低。

With the development of digitalization, digital images, as critical carriers of information, are widely used across various sectors, including social media, healthcare, finance, and the military. While digital images offer convenience in transmission and storage, their security challenges have become increasingly prominent. In response to these security challenges, China has enacted The Data Security Law, The Personal Information Protection Law, and the Cryptography Law of the People’s Republic of China. Image encryption technology, as an effective measure for safeguarding privacy, employs specific encryption algorithms to reversibly convert digital images into unreadable or unusable forms, ensuring comprehensive protection for digital images during storage and transmission.
JPEG images, as one of the most widely used formats today, present a particularly valuable context for discussing privacy and security concerns. Compared with uncompressed images, JPEG encryption must address not only the security of the encrypted image but also issues of file size and format compatibility, given its strict encoding format. In addition, it is essential to consider users’ requirements for the usability of encrypted images, such as quick identification and previewing. Balancing JPEG image security and usability while achieving low file expansion and format compatibility is undoubtedly a significant challenge. Existing JPEG image encryption algorithms perform well in terms of format compatibility and file size expansion. However, they preserve image block feature values that construct adaptive encryption key, along with other block feature values, leaving the encrypted images vulnerable to chosen-plaintext attacks. To enhance the privacy protection capabilities of JPEG image encryption algorithms, this paper first conducts a security analysis on the JPEG encryption algorithm that preserves block feature values, and then designs three different encryption schemes: the first is to change the block feature of key construction; the second is an encryption algorithm that improves the change rate of block feature values; the third is to propose the JPEG thumbnail preserving encryption (JPEG-TPE) based on the second algorithm, to cater to the needs of some users for the usability of encrypted images. The specific research contents are as follows:
1. Analyzed the security of JPEG image encryption algorithm with block feature preservation and proposed” block eigenvalue change rate” metric. The unchanged block feature values used to construct the encryption key allows the attacker to easily construct a plaintext image with the same key as the image to be attacked, which is a prerequisite for CPA; The unchanged other block feature values contribute to the rapid estimation of the block permutation sequence and can reveal the sketch of original image after block permutation sequence is estimated right. Experimental results demonstrate that the JPEG image encryption algorithm preserving block features which used to construct the adaptive encryption key and other block feature values are vulnerable to CPA, allowing attackers to obtain sketch and/or the original image. Furthermore, by analyzing the linear relationship between the estimation accuracy of the block permutation sequence and the change rate of block features, we propose the metric ”change rate of block features” for the JPEG image adaptive key encryption scheme with combining block permutation with other encryption steps to measure the ability to resistant to CPA attack.
2. JPEG image encryption with changed block feature for constructing adaptive encryption key. Utilizing the existing Reversible data hiding (RDH) scheme based on VLC mapping, the block feature used to construct the encryption key can be changed by self-embedding RSV (run/size, value), which can destroy the prerequisite of CPA attack without increasing the burden of key management. After analyzing the monotonic relationship between embedding capacity and file size in existing VLC-based RDH schemes, the maximum allowable RSV embedding threshold for reducing file size is determined. A random selection of RSVs, not exceeding this threshold, is self-embedded into the DC and AC encrypted image to generate the encrypted image. Experimental results show that the proposed algorithm reduces the probability of accurately reconstructing adaptive keys directly from encrypted images to nearly zero, even when the file size of encrypted image is reduced by 1,000 bits compared to the original image.
3. JPEG image encryption with improved change rate of block feature value. To improve the block feature value change rate without increasing the file size, the concept of AC coefficients group with undivided RSV(ACG-URSV) is proposed for the first time. Additionally, a global scrambling scheme for ACG-URSV is designed. Since different ACG-URSVs contain different number, different run length, and different values of nonzero AC, encrypting ACG-URSVs of the same length can change the different block feature values. Furthermore, to reduce file size expansion, an encryption scheme is designed to rotate the DC coefficient based on DC difference after mapping to the same category. Experimental results show that the proposed algorithm increases the block feature value change rate by at least 20%, while the average file size decreases by 2.036%.
4. JPEG thumbnail preserving encryption (JPEG-TPE) with improved change rate of block feature value. To address the issues of low feature value change rate of existing JPEG-TPE schemes and reduction in the number of encryption elements in ACG-URSV global permutation encryption (because of fixed frequencies), an adaptive ACG-URSV global permutation is designed. To preserve the thumbnail while minimizing file size expansion, the DC difference values rather than the DC coefficient values are redistribution with unchanged sum. Experimental results show that the thumbnail quality of the proposed algorithm is basically the same as that of the existing schemes, but the average number of permutation elements increases by 1809, the file expansion of the image is reduced by 14.77%, the block feature value change rate increases by 38.28%, and this improvement does not diminish with increasing encrypted image block size.