Pixel-Batched Homomorphic Encryption for Secure Outsourced Image Convolution
Abstract
Secure outsourced image processing requires a workflow where a data owner wants to simultaneously keep her imagery private while providing it to another party for storage and computation. A key challenge involves allowing the other party to perform meaningful computations on behalf of the data owner. Over the last ten years, homomorphic encryption research addressing this challenge has blossomed. Owing to the fundamental nature of convolution in image processing, we address the problem of secure outsourced image convolution. We start by developing an approach based on a straight-forward application of the Paillier cryptosystem where each pixel is encrypted separately. We improve upon this approach by batching vectors of pixels before encryption (an idea applied by others to different image processing algorithms). We carried out a series of experiments and observed the approach involving batching to require one to two orders of magnitude less computation time.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 19, 2020
- Accession Number
- AD1156987
Entities
People
- Adrian V Mariano
- Chris R Giannella
- James H Tanis
Organizations
- MITRE Corporation