Cryptology ePrint Archive: Report 2017/857

Image Classification using non-linear Support Vector Machines on Encrypted Data

Anthony Barnett and Jay Santokhi and Michael Simpson and Nigel P. Smart and Charlie Stainton-Bygrave and Srnivas Vivek and Adrian Waller

Abstract: In image processing, algorithms for object classification are typically based around machine learning. From the algorithm developer's perspective, these can involve a considerable amount of effort and expertise to develop, which makes them commercially valuable. On the other hand, other parties may want to make use of these algorithms to classify their images, while protecting the privacy of their data. In this paper, we show how non-linear Support Vector Machines (SVMs) can be practically used for image classification on data encrypted with a Somewhat Homomorphic Encryption (SHE) scheme. Previous work has shown how an SVM with a linear kernel can be computed on encrypted data, but this only has limited applicability. By enabling SVMs with polynomial kernels, a much larger class of applications are possible with more accuracy in classification results.

Category / Keywords: cryptographic protocols /

Date: received 5 Sep 2017

Contact author: nigel at cs bris ac uk, sv venkatesh at bristol ac uk, Anthony Barnett at uk thalesgroup com, Adrian Waller at uk thalesgroup com

Available format(s): PDF | BibTeX Citation

Version: 20170909:213326 (All versions of this report)

Short URL:

[ Cryptology ePrint archive ]