PRIVACY-PRESERVING ALGORITHMS FOR MULTIMEDIA PROCESSING OVER CLOUD

Abstract

The emerging advancement of cloud-computing has attracted users worldwide to utilize storage and processing services in a hassle-free manner. These services require low maintenance costs, access exibility, scalability, data-loss prevention, and mobility to smart gadgets. A third party maintains the cloud servers; therefore, these cloud services have drawbacks requiring the user's raw data processing. The data, especially multimedia, contains personal and con dential information whose leakage may create personal problems in users' lives. Therefore, when moving the data to the cloud server, data privacy is always the user's concern as third parties may misuse their data for monetary and other related bene ts. In the past decade, researchers have addressed data transmission and storage problems. However, secure data processing is still an open problem, which requires e cient and robust privacy-preserving schemes with a high-level of mathematically proved security aspects. This thesis focuses on developing cloud-based systems for secure computations to solve ve major real-world problems over the multimedia dataset. These problems include a cloud-based privacy-preserving image inpainting model, privacy-preserving exemplar-based object-removal over the cloud, privacy-preserving image recognition system for cloud infrastructure, privacy-preserving cloud-based deep learning system for sign language recognition (SLR), and privacy-preserving model for skin lesion segmentation as a medical imaging application. Besides the secure computations, we propose two novel image encryption schemes that can be utilized for processing the image and video datasets over the cloud platforms. The complete thesis work is divided into seven chapters.