Abstract:

More and more enterprise and individual users choose to store a large number of image files in the cloud server, and provide image retrieval and sharing functions. In order to protect the information of the stored important images from being stolen, image files are stored in the cloud server in encrypted form, which brings challenges to the image retrieval operation. The traditional plaintext retrieval schemes are no longer applicable, and how to guarantee the retrieval efficiency and accuracy of a large number of ciphertext images is also an important issue. According to the above problems, a ciphertext image retrieval scheme based on target detection in cloud environment is proposed. In the scheme, a target detection model named Faster R-CNN which is based on deep learning is used to extract keyword sets and feature vectors from images accurately, then the image sets are classified coarsely using the keyword sets, and the keywords are encrypted to construct secure indexes using the multiple linear mapping, so as to retrieve the matching image sets efficiently, at last the feature vectors of images are matched precisely to achieve the fine classification of images, retrieving the final images. The security analysis and performance evaluation show that the scheme has high security, retrieval efficiency and accuracy.

Key words: cloud computing, ciphertext image retrieval, object detection, multiple linear mapping, security index

摘要：

越来越多的企业和个人用户选择将大量的图像文件存储在云服务器中，并提供图像的检索和共享功能。为了保障所存储的重要图像信息不被窃取，图像文件以加密的形式存储在云服务器中，这给图像的检索操作带来了挑战。传统的明文检索方案已经无法适用，并且如何保证大量密文图像数据的检索效率和精确度也是一个重要问题。针对上述问题，提出了一种云环境中基于目标检测的密文图像检索方案，利用基于深度学习的目标检测模型Faster R-CNN对图像精确提取关键词集合和特征向量，使用关键词集合对图像集合粗分类，使用多重线性映射对关键词加密并构建安全索引，以高效检索出匹配的图像集合，再对图像特征向量精确匹配，实现图像的细分类，以检索出最终的图像。安全性分析和性能评估表明该方案具有高安全性、检索效率和精确度。

关键词: 云计算, 密文图像检索, 目标检测, 多重线性映射, 安全索引