Logo image
Back
Secure Feature Extraction in Computational Vision Using Fully Homomorphic Encryption
Conference proceeding

Secure Feature Extraction in Computational Vision Using Fully Homomorphic Encryption

Thomas Shortell and Ali Shokoufandeh
PROCEEDINGS OF THE FUTURE TECHNOLOGIES CONFERENCE (FTC) 2018, VOL 2, v 881
01 Jan 2019
DOI: https://doi.org/10.1007/978-3-030-02683-7_15

Abstract

Computer Science Computer Science, Artificial Intelligence Science & Technology Technology
Cloud computing is an important part of today's critical infrastructure because of its integration to all aspects of modern day services. One such set of services are those relying on computational vision systems. The reliability and quality of service for such systems depend on the accuracy of extracted visual features. As such the security of extracting these features is a single point of failure in a distributed computing architecture. This research investigates implementing Speeded-Up Robust Features (SURF) and Histograms of Oriented Gradients (HOG) in the encrypted domain using Fully Homomorphic Encryption (FHE). This provides a method for a user to reduce their risk in offloading processing to a computationally powerful cloud resource. A framework is developed for two different numerical format systems to support real numbers in the FHE realm. Bounding the error introduced in the framework is also investigated to enable the system to provide desired numerical accuracy. Results of implementing the framework in SURF and HOG shows these feature extraction algorithms can be computed securely in the encrypted domain.

Metrics

22 Record Views
1 citations in Scopus

Details

InCites Highlights

Data related to this publication, from InCites Benchmarking & Analytics tool:

Web of Science research areas
Computer Science, Artificial Intelligence
Logo image