Privacy-Preserving Krawtchouk Moment feature extraction over encrypted image data

Abstract

Resource-constrained users outsource the massive image data to the cloud to reduce storage and computation overhead locally, but security and privacy concerns seriously hinder the applications of outsourced image processing services. Besides, existing image processing solutions in the encrypted domain still bring high computation overhead, and even lead to characteristic loss. To this end, we propose a Privacy-Preserving Krawtchouk Moment (PPKM) feature extraction framework over encrypted image data by utilizing the Paillier cryptosystem. First, a mathematical framework for PPKM implementation and image reconstruction is presented in the encrypted domain. Then, the detailed expanding factor and upper bound analysis shows that plaintext Krawtchouk moment and plaintext image reconstruction can be realized over encrypted image with PPKM. Furthermore, the computation complexity of PPKM can be significantly reduced with the block-based parallel algorithm. Experimental results verify that the PPKM is feasible and acceptable in practice in terms of image reconstruction capability and image recognition accuracy.