The vigorous development of the Internet of Things and cloud computing is driving resource-limited smart devices to outsource large-scale images to cloud servers for storage and retrieval. Privacy-preserving image retrieval addresses the threat of data privacy leakage without affecting the searchability of images. Existing privacy-preserving retrieval schemes use the approximate nearest neighbor search to improve the retrieval efficiency of large-scale images on the cloud server. However, these schemes suffer from reduced retrieval accuracy, difficulties in constructing encrypted index structures, and a lack of result verification support. To tackle these problems, we propose a verifiable privacy-preserving retrieval scheme for large-scale images (VPIRL) in cloud servers. We use learning with errors (LWE) theory to protect image features, achieving distance and angle preservation between encrypted features. This enables the cloud server to construct an encrypted satellite system graph for efficient and accurate retrieval of large-scale images. We also propose a privacy-preserving data verification method based on the Merkle Hash Tree and cuckoo hash to detect dishonest behaviors of the cloud server and verify the correctness and completeness of the approximate nearest neighbor retrieval results. Experimental results show that this scheme achieves retrieval and verification in milliseconds for millions of images, confirming its practicality for large-scale image retrieval.

Cloud-based indoor positioning services have advantages over non-cloud methods but also confront serious privacy concerns. Existing privacy-preserving schemes are designed for conventional two-entity localization models thus not applicable to the cloud-based indoor positioning services involving three entities. In addition, these methods incur high computational and communication overhead. To tackle these issues, we proposed a privacy-preserving indoor positioning scheme for WiFi localization based on Inner Product Encryption in a cloud environment. A bloom filter constructed with Locality Sensitive Hashing was designed to map WiFi fingerprints from Euclidean to inner product space with the distance relationships maintained for converting the location estimation to inner product calculations. Inner Product Encryption protects the user’s fingerprint and database information held by the positioning service provider. Fingerprint similarity as determined by the inner product is decrypted on the cloud to retrieve the closest encrypted location coordinates for users. In addition, a retrieval structure based on Hierarchical Navigable Small World graph was designed to improve efficiency. Theoretical analysis and experimental results demonstrate that the scheme has low computational and communication overhead while ensuring security and not significantly degrading the localization accuracy. Moreover, the overhead does not increase significantly with database size thus this approach is highly scalable.

Cellular traffic prediction will play a key role in the deployment of future smart cities. Although the current traffic prediction methods based on deep learning show better performance than traditional prediction methods, they still have the following problems: (1) In spatial domain, the correlations between cellular traffic features cannot be captured accurately in non-local (including “geographic adjacency” and long-distance) spatial areas. (2) In temporal domain, the correlation of different time-grained features is failed to consider. To address these problems, a deep learning method considering dynamic non-local spatial correlation, self-attention, and correlation of spatio-temporal feature fusion is proposed. In spatial domain, our method can accurately capture the spatial correlation and highlight the contribution of more relevant traffic in the non-local area by designing a NLG-NLAM model. In temporal domain, the correlations of time-periodic features with different granularities are considered to clarify the key roles of different periodic features and eliminate the influence of irrelevant cellular traffic features on the prediction by designing a calibration layer. Experimental results indicate that the proposed method shows better performance than other mainstream prediction methods on three real-world cellular traffic datasets.

Users outsource images to edge servers physically closer to their location for real time applications because of the low latency and low transmission overhead. Outsourcing to these edge servers however, increases the risks to data privacy. Almost all existing privacy preserving image retrieval schemes utilize a single cloud server to execute retrieval tasks and provide centralized image retrieval but at high computational costs, thus are not suitable for the distributed edge environments with limited computing resources. We propose a lightweight privacy-preserving multi-source image retrieval scheme adapted specifically for the distributed edge environment. We apply high efficiency orthogonal decomposition and learning with errors (LWE) strategy to encrypt image features and construct cipher indexes and trapdoors, guaranteeing the security of the data, while reducing computational costs. The orthogonality of data ensures that the accuracy of retrieval results is not compromised by the random numbers used in the scheme. In addition, the proxy re-encryption technology is adopted to support the retrieval of multi-source images encrypted by unique data owners with different keys. A detailed performance analysis and comprehensive experiments demonstrate that our scheme guarantees data security with very high retrieval accuracy and a low computational burden, consistent with the demands of edge environments.

Convolutional neural networks (CNN) is a popular architecture in machine learning for its predictive power, notably in computer vision and medical image analysis. Its great predictive power requires extensive computation, which encourages model owners to host the prediction service in a cloud platform. This article proposes a CNN prediction scheme that preserves privacy in the outsourced setting, i.e., the model-hosting server cannot learn the query, (intermediate) results, and the model. Similar to SecureML (S&P’17), a representative work that provides model privacy, we employ two non-colluding servers with secret sharing and triplet generation to minimize the usage of heavyweight cryptography. We made the following optimizations for both overall latency and accuracy. 1) We adopt asynchronous computation and SIMD for offline triplet generation and parallelizable online computation. 2) As MiniONN (CCS’17) and its improvement by the generic EzPC compiler (EuroS&P’19), we use a garbled circuit for the non-polynomial ReLU activation to keep the same accuracy as the underlying network (instead of approximating it in SecureML prediction). 3) For the pooling in CNN, we employ (linear) average-pooling, which achieves almost the same accuracy as the (non-linear, and hence less efficient) max-pooling exhibited by MiniONN and EzPC. Considering both offline and online costs, our experiments on the MNIST dataset show a latency reduction of$122\times$,$14.63\times$, and$36.69\times$compared to SecureML, MiniONN, and EzPC; and a reduction of communication costs by$1.09\times$,$36.69\times$, and$31.32\times$, respectively. On the CIFAR dataset, our scheme achieves a lower latency by$7.14\times$and$3.48\times$and lower communication costs by$13.88\times$and$77.46\times$when compared with MiniONN and EzPC, respectively.