Exploring Possibilities of BFV-based Homomorphic Encryption for Privacy-Preserving Image Processing

We explore how homomorphic encryption (HE) can be used for secure medical image processing. While the CKKS scheme is the obvious choice because it supports floating-point numbers, it has drawbacks like larger ciphertexts and slower performance compared to the BFV and BGV schemes. BFV and BGV are faster and more efficient, but they can only handle integers, which limits their ability to perform more advanced image processing tasks. To overcome this, we modify the BFV scheme to work with fixed-point arithmetic, allowing it to handle basic real-number calculations. We tested this approach on simple image processing tasks like brightness enhancement, greyscaling, and sharpening filter, showing it can securely process images in the encrypted domain.