TWO-LAYER FRAGILE WATERMARKING METHOD FOR ENHANCED TAMPERING LOCALISATION Sergio Bravo-Solorio ∗† and Asoke K. Nandi Department of Electrical Engg. & Electronics. The University of Liverpool Brownlow Hill, Liverpool, L69 3GJ, UK; a.nandi@liverpool.ac.uk, S.Bravo-Solorio@warwick.ac.uk ABSTRACT This paper presents a new fragile watermarking method, whereby a secure block-wise and an interlaced watermarking mechanisms are hierarchically structured to provide higher tampering localisation capabilities. The block-wise method provides security against con- ventional distortions, cropping, as well as sophisticated attacks, whereas the interlaced scheme is aimed at enhancing the locali- sation accuracy achieved by the block-wise method. Results are presented to illustrate the improved localisation capabilities of the proposed method in comparison with some five existing watermark- ing schemes. Index Terms— Fragile watermarking, tampering localisation, authentication. 1. INTRODUCTION The availability of image editing software equipped with sophisti- cated tools has sparked serious concerns about the integrity of digital images, especially in application domains where sensitive informa- tion is handled – e.g. law enforcement applications. Fragile water- marking describe techniques to embed information imperceptibly – i.e. a watermark – in digital images, so that detected changes in the watermark indirectly expose manipulations in the host image [1]. Fragile watermarking systems can produce binary (yes/no) an- swers to state whether or not a host image has been tampered with. Nonetheless, the fact that a watermark undergoes the same transfor- mation as the host image opens up the possibility of identifying the tampered region, while verifying the remainder of the image. This important feature is commonly referred to as tampering localisa- tion (or simply localisation). Identifying the altered pixels correctly could be useful to evaluate whether the semantic meaning of an im- age has been affected significantly. Furthermore, depending on the nature of the distortion, it may be possible to restore the altered con- tent by means of denoising or inpainting techniques, e.g. [2]. Wong [3] proposed a scheme, whereby a message authentication code (MAC) is independently generated from the seven most signifi- cant bit-planes (MSBPs) of every non-overlapping pixel-block in the image. Then, the least significant bit-plane (LSBP) of each block is replaced with the MAC derived from the block itself. This method provides an effective localisation, but is susceptible to vector quan- tisation (VQ) attacks [4]. That is, the blocks of a host image can be swapped without being noticed or, even worse, blocks copied from a set of images watermarked with the same key can be assembled together to form a completely new image that would go undetected. ∗ Sergio Bravo-Solorio is supported by the National Council of Science and Technology (CONACyT) of Mexico. † Sergio Bravo-Solorio is now with the Department of Computer Sciences at the University of Warwick. To localise distortions, while effectively thwarting VQ attacks, Fridrich [5] proposed to make the MAC dependent on a block-index and a unique image index. Lin et al. [6] used the six MSBPs of every pixel-block to encode a MAC, which is embedded in the two LS- BPs of the subsequent block in a block-mapping sequence generated pseudo-randomly. A hierarchical mechanism is adopted to localise and even recover the altered blocks. However, the correlation be- tween the blocks can be estimated to perform successful attacks [7]. In [8], a sparse set of wavelet coefficients are watermarked in accor- dance with a contextual non-deterministic dependence mechanism, which involves all the wavelet coefficients across the wavelet decom- position levels. The aim is to protect all the pixels without altering all the wavelet coefficients. He et al. [9] proposed to map every wavelet coefficient, in the coarser sub-band, to a 4-bit code, which is then embedded in the 2 × 2 pixel-block corresponding to another wavelet coefficient selected pseudo-randomly. At the receiver side, possibly distorted codes are localised in a bitmap, and then a post- processing mechanism is adopted to remove isolated blocks. In [10], the MAC, derived from the five MSBs of every single pixel and a pseudo-random code, is embedded into the three LSBPs of the im- age. At the receiver end, two distributions, corresponding to the altered and genuine pixels, are employed to localise pixels corrupted in their five MSBs. This method manages to localise altered pixels accurately only if the altered area is not too extensive. In [11], a se- cret key is used to generate a circular block-mapping sequence. The MAC computed for a block is embedded in the subsequent block in the sequence. One of the four least significant bits (LSBs) in every pixel is pseudo-randomly selected to allocate a bit of the MAC. In this paper, a new method is presented to provide enhanced tampering localisation. The proposed method relies on two water- mark layers. One is embedded by means of a secure block-wise method resilient to cropping. A second watermark, which is spread over the image following an interlaced arrangement, is used to re- fine the localisation achieved at the first layer. Results demonstrate that the proposed two-layer (TL) fragile watermarking method out- performs five state-of-the-art methods, in terms of localisation per- formance. The TL method is detailed in Section 2, while Section 3 elaborates on the embedding distortion and the security aspects of the scheme. Some results are presented in Section 4 and the paper is concluded in Section 5. 2. PROPOSED METHOD 2.1. Embedding process Consider an n1×n2 grey-scale image X divided into non-overlapping blocks of m × m pixels, where m =2c1 for some c1 ∈ N; addi- tionally, it will be assumed that n1 = c2m and n2 = c3m, for any c2,c3 ∈ N. Let Xp be the p-th block in X, for p =1,...,mbw = 2245 978-1-4673-0046-9/12/$26.00 ©2012 IEEE ICASSP 2012