Welcome to the resource topic for 2024/559
Title:
Convolution-Friendly Image Compression in FHE
Authors: Axel Mertens, Georgio Nicolas, Sergi Rovira
Abstract:Fully Homomorphic Encryption (FHE) is a powerful tool that brings privacy and security to all sorts of applications by allowing us to perform additions and multiplications directly on ciphertexts without the need of the secret key.
Some applications of FHE that were previously overlooked but have recently been gaining traction are data compression and image processing.
Practically, FHE enables applications such as private satellite searching,
private object recognition, or even encrypted video editing.
We propose a practical FHE-friendly image compression and processing pipeline where an image can be compressed and encrypted on the client-side, sent to a server which decompresses it homomorphically and then performs image processing in the encrypted domain before returning the encrypted result to the client.
Inspired by JPEG, our pipeline also relies on discrete cosine transforms
and quantization to simplify the representation of an image in the frequency domain, making it possible to effectively use a compression algorithm.
This pipeline is designed to be compatible with existing image-processing techniques in FHE, such as pixel-wise processing and convolutional filters.
Using this technique, a high-definition (1024\times1024) image can be homomorphically decompressed, processed with a convolutional filter and re-compressed in under $24.7$s, while using ~8GB memory.
ePrint: https://eprint.iacr.org/2024/559
See all topics related to this paper.
Feel free to post resources that are related to this paper below.
Example resources include: implementations, explanation materials, talks, slides, links to previous discussions on other websites.
For more information, see the rules for Resource Topics .