一种智能电网敏感图像可检索的属性基加密方案

Retrievable Attribute-Based Encryption Scheme for Sensitive Images of Smart Grid

针对智能电网环境中敏感图像数据面临的被窃取、被滥用等安全威胁,及现有密文图像检索方案难以兼具高检索速度与高检索精度的问题,提出了一种面向智能电网环境的基于多特征融合的敏感图像可检索属性基加密方案。利用神经网络模型提取深度图像特征,并分别提取包括词袋特征、颜色空间特征及方向梯度直方图特征在内的3种传统图像特征,采用主成分分析法将这4种特征进行融合与降维,以增强特征向量的表示信息;基于ρ稳定局部敏感哈希函数和改进后的安全近邻算法对敏感图像集和查询图像分别构建安全索引和搜索陷门,通过对安全索引与搜索陷门进行相似度计算得到密文图像检索结果;将密文策略属性基加密机制与可搜索加密进行结合,只有符合访问策略的用户才能得到明文图像检索结果。理论分析和实验分析表明,在真实的电力数据集中,所提方案的密钥生成时间、索引生成时间及检索时间均为毫秒级,检索精度为88.6%。3种对比方案的实验结果表明,所提方案在检索效率与检索精度中的平衡性最好,深度特征的融合能够将3种对比方案的检索精度提高5.7%~11.4%。安全性分析表明,所提方案可抵抗已知密文攻击、已知背景攻击和抗串谋攻击。

Aiming at the threats to security that sensitive images of smart grid are stolen and abused,and at the problem that the existing ciphertext image retrieval solutions are difficult to achieve both high retrieval rate and accuracy,a retrievable attribute-based encryption scheme for sensitive images of smart grid with multi-feature fusion is proposed.The scheme extracts deep image features through the Dense convolutional network,and respectively extracts three traditional image features including bag-of-words features,colorspace features,and histogram of oriented gradients features.The principal component analysis algorithm is used to fuse and reduce the dimensions of four features to enhance the representative information of the feature vector.Based on theρstable locality-sensitive hashing function and the improved secure nearest neighbor algorithm,the security index and the trapdoor are constructed for the sensitive image dataset and the query image,respectively.The ciphertext image retrieval result is obtained by calculating the similarity between the security index and the trapdoor.The ciphertext-policy attribute-based encryption mechanism is combined with searchable encryption,and only the users who meet the access strategy can obtain the plaintext image retrieval results.Theoretical analysis and experimental analysis indicate that in the real power dataset,the key generation time,index generation time and retrieval time of the proposed scheme are all in milliseconds,and the retrieval accuracy is 88.6%.Experimental results of the three comparative schemes indicate that the proposed scheme has the best balance between retrieval efficiency and retrieval accuracy.The fusion of deep features can increase the retrieval accuracy of the three comparative schemes by 5.7%-11.4%.The security analysis shows that the proposed solution can resist known ciphertext attacks,known background attacks and collusion attacks.