Abstract:
Surveillance data images of crowded places have become increasingly important in today’s world for a variety of reasons. Cloud-based image searching schemes leverage distributed computing power to search large amounts of data quickly and accurately, allowing users to easily find relevant images based on their search queries. Multiple schemes have been proposed but there remain issues related to the security and privacy of data owners as well as the revelation of search and access patterns. Moreover, existing homomorphic techniques for encrypted data are inefficient enough to be deployed in a real-world environment. A novel scheme for similar image searching based on homomorphic encryption is proposed for cloudconnected devices. The scheme is based on probabilistic trapdoors and ensures the privacy preservation of image data. The scheme also provides search pattern security and is CPA-secure in an adversarial model. The scheme also fulfills the property of Query-Trapdoor and Trapdoor-Image Indistinguishability. Implementation and testing of the proposed scheme are carried out over a real-world data set in order to assess its security and performance in terms of complexity, computation, and storage overheads.
