A Cloud-Based Approach for Privacy-Preserving Medical Image Retrieval: Leveraging Local Features and PCA in Two Efficient Steps

Abstract

The major challenges of content-based image retrieval are searching, ranking, and retrieving in secure mode. In this article, we propose a two-step security approach for each database image, utilizing encryption mechanisms and cloud technology. In this approach, the watermark-embedded encrypted images, along with the efficient feature vector database and the list of authenticated users, will be stored in the cloud. During retrieval, upon verification of the authenticated user in the cloud, the encrypted images will be accessed as the first step in the security process. In the second step, after further user verification, a key provided by the image owner will allow the watermark to be retrieved, enabling the end user to access the original image. Feature vector dataset for all dataset images is constructed using a dominant local pattern named RDEBP combined with PCA. In this two-step security approach for database images, the watermark-embedded system is designed to provide security from unauthenticated users and also from duplicating the images after the first stage of retrieval. Customized watermark bits are embedded into each block of encrypted images before they are stored in the cloud. Consequently, if any unauthorized duplicate image is found, then the watermark-extraction module traces the source and identifies the user, who is responsible for circulating the image. The significance of the proposed method compared to its watermark encryption accuracy by varying the size of blocks has been verified. And also verified its retrieval accuracy over many existing methods, showcased in terms of mean average precision and recall. Experimental trials and security analyses confirm that the proposed approach is both robust and efficient, ensuring a secure and reliable system.