Medical Images Authentication through Repetitive Index Modulation Based Watermarking

Introduction

Due to privacy concerns and authentication needs, many digital watermarking schemes (Bao et al, 2005; Coatrieux et al, 2001; Guo & Zhuang, 2007; Kong & Feng, 2001; Osborne et al, 2004; Planitz & Maeder; 2005; Zhou et al, 2001) have been proposed to embed authentication data into the contents of medical images. Methods proposed in the literature can be broadly classified into two categories: spatial domain watermarking (Bao et al, 2005; Coatrieux et al, 2001; Kong & Feng, 2001) and transform domain watermarking (Wakatani, 2002; Lie et al, 2003; Osborne et al, 2004). Most transform domain watermarking methods are designed to work with lossy compression standards, such as JPEG and JPEG 2000. The main concern surrounding this type of watermarking schemes is that in most cases lossy compression is not allowed to be applied to medical images, thus restricting their applicability. On the other hand, most spatial domain embedding methods are developed for the applications in which no lossy compression is expected. Many spatial domain embedding methods (Bao et al, 2005; Coatrieux et al, 2001; Kong & Feng, 2001) require that the least significant bits (LSBs) of the image pixels be replaced with the authentication codes or watermarks. Although the distortion due to this kind of “destructive” watermark embedding is usually visually insignificant, medical images with watermarks embedded with this type of irreversible watermarking schemes may not be acceptable as feasible evidence in the court of law, should medical disputes occur. Many reversible data hiding schemes (Li, 2005; Thodi & Rodriguez, 2007), although not specifically proposed for the purpose of medical image authentication, have been developed to facilitate reversible data hiding, in which the original images can be recovered after the hidden data is extracted from the watermarked images. A reversible watermarking scheme specifically developed for authenticating medical data has been proposed in (Kong & Feng, 2001). The common problem with these reversible data hiding schemes is that, apart from the actual payload (i.e., the watermark, secret data, authentication codes, etc), side information for reconstructing the exact original image has to be embedded as well. The side information wastes limited embedding capacity and is usually the compressed form of the location map of the original data that is expected to be affected by the embedding process. The waste of embedding capacity reduces the authentication power of the scheme and the resolution of tamper localization, as explained in (Li & Yuan, 2006). Moreover, authentication schemes are also expected to be resistant against attacks, such as the Holliman-Memon counterfeiting attack (Holliman & Memon, 2000), the birthday attack (Stallings, 1998) and the transplantation attack (Barreto et al, 2002), by involving the contents in the watermarking process in a non-deterministic manner (Kim et al, 2008; Li & Yuan, 2006). Therefore schemes with high payload, high resolution of tamper localization, high security and zero distortion to the ROI are desirable.