摘要：

随着云计算和云存储场景中图像安全和用户隐私需求日益增加, 密文图像可逆数据隐藏技术备受关注。相较于加密前预留空间(RRBE)的方法, 传统的加密后腾出空间(VRAE)技术通常在加密过程中破坏了原始图像像素间的相关性, 导致嵌入率降低, 从而限制其应用广泛性。为提升VRAE的嵌入率同时确保图像安全, 提出一种可用于云计算环境的基于四叉树加密和自适应预测误差编码的加密图像可逆数据隐藏方案。首先, 采用基于四叉树的分区置乱加密算法, 在确保图像安全性的同时保证块内像素的相关性, 并利用中值边缘预测器获取块内像素值的预测误差; 其次, 对预测误差的数值位进行自适应块编码, 根据块的大小采用不同的编码方法, 从而有效压缩数据并腾出空间供数据嵌入。实验结果表明, 与现有的密文域可逆数据隐藏方案相比, 该方案更有效地利用了像素间的相关性, 提高了秘密信息的嵌入能力, 在BOSSBase和BOWS-2数据集上平均嵌入率分别达到3.332 bit/pixel和3.289 bit/pixel, 比现有先进的VRAE方法分别提高0.117 bit/pixel和0.175 bit/pixel。

关键词: 可逆数据隐藏, 图像安全, 自适应块编码, 预测误差, 四叉树加密

Abstract:

With the increasing demand for image security and user privacy in cloud computing and storage scenarios, ciphertext image reversible data hiding techniques have garnered significant attention. Compared to the Reserve Room Before Encryption (RRBE) method, the traditional Vacate Room After Encryption (VRAE) technology typically destroys the correlation between the pixels of the original image during the encryption process. This results in a low embedding rate, thereby limiting its broader application. To improve the embedding rate of VRAE and ensure image security, this study proposes a reversible data hiding scheme for encrypted images that employs quadtree-based encryption and adaptive prediction error encoding, specifically designed for cloud computing environments. The scheme uses a quadtree-based partition scrambling algorithm to preserve intra-block pixel correlations while ensuring image security. The median edge predictor is used to obtain the prediction error of pixel values within the block. It quantitatively encodes prediction errors within blocks using adaptive block encoding techniques tailored to block sizes, effectively compressing data and creating embedding space. The results demonstrate that the proposed scheme better utilizes pixel correlations and achieves higher embedding capacity for confidential information compared to existing ciphertext-domain reversible data hiding methods. On the BOSSBase and BOWS-2 datasets, the average embedding rates reach 3.332 bit/pixel and 3.289 bit/pixel, respectively, surpassing those of state-of-the-art VRAE methods by 0.117 bit/pixel and 0.175 bit/pixel.

Key words: reversible data hiding, image security, adaptive block encoding, prediction error, quadtree encryption