The rapid growth of digital images motivates organizations and individuals to outsource image storage and computation to the cloud. However, the defenseless upload will raise the risk of privacy leakage while the simple encryption would impede the efficient usage of data. In this paper, we propose a privacy-preserving image retrieval scheme, in which the images are encrypted but similar images to a query can be efficiently retrieved from the encrypted images. Specifically, the image content is protected by big-block permutation, $3 \times 3$ block permutation within big-blocks, pixel permutation within $3 \times 3$ blocks, and polyalphabetic cipher. The use of polyalphabetic cipher improves security and causes no degradation in terms of retrieval accuracy as the substitution tables are generated by the order-preserving encryption. In this way, secure Local Binary Pattern (LBP) features can be directly extracted as the local features from the encrypted big-blocks, which is efficient as there is no communication between the cloud server and image owners to do so. The secure local LBP features are used to generate the feature vector for each image by the bag-of-words model. Finally, the similarity among the encrypted images is measured by the Manhattan distance of such feature vectors. The security analysis and experimental results demonstrate that the proposed scheme outperforms the main existing schemes in terms of security and retrieval accuracy.

中文翻译：

---

云计算中使用安全本地二进制模式的隐私保护图像检索方案

数字图像的快速增长促使组织和个人将图像存储和计算外包给云。但是，无防御的上载会增加隐私泄露的风险，而简单的加密会阻碍数据的有效使用。在本文中，我们提出了一种隐私保护的图像检索方案，该方案对图像进行加密，但是可以从加密的图像中高效检索与查询相似的图像。具体来说，图片内容受大块排列保护，$ 3 \时间3 $ 大块内的块排列，内大块的像素排列 $ 3 \时间3 $块和多字母密码。多字母密码的使用提高了安全性，并且由于替换表是通过保留顺序的加密生成的，因此在检索准确性方面不会降低。这样，可以直接从加密的大块中提取安全的本地二进制模式（LBP）功能作为本地功能，这是有效的，因为云服务器和映像所有者之间没有通信。安全局部LBP特征用于通过词袋模型为每个图像生成特征向量。最后，通过这种特征向量的曼哈顿距离来测量加密图像之间的相似性。安全性分析和实验结果表明，该方案在安全性和检索准确性方面均优于现有的主要方案。