Zero-Watermarking for Face Image Protection in Database

Abstract

A face image is usually registered in a database as an important piece of personal information. To protect the copyright and privacy of face images, we propose a new zero-watermarking scheme based on image morphing for face images, in which morphed images are registered as zero-watermark. The symmetry of the human eye is used to preprocess the test face image, and an improved image fusion method is used to eliminate ghosting in registered zero-watermark images. In addition, a new image matching method called ＂combined moment invariants＂ is proposed to match zero-watermark images, which is based on Hu moment invariants and Krawtchouk moment invariants. Experimental results show the proposed scheme not only protect the privacy of face images in database, but also shows better robustness to noise, filter, JPEG compression, rotation and other attacks.