Two key estimation techniques for the broken arrows watermarking scheme

Abstract

This paper presents two different key estimation attacks targeted for the image watermarking system proposed for the BOWS-2 contest. Ten thousands images are used in order to estimate the secret key and remove the watermark while minimizing the distortion. Two different techniques are proposed. The first one combines a regression-based denoising process to filter out the component of the original images and a clustering algorithm to compute the different components of the key. The second attack is based on an inline subspace estimation algorithm, which estimates the subspace associated with the secret key without computing eigen decomposition. The key components are then estimated using Independent Component Analysis and a strategy designed to leave efficiently the detection region is presented. On six test images, the two attacks are able to remove the mark with very small distortions (between 41.8 dB and 49 dB). Categories and Subject Descriptors General TermsSecurity, Algorithms Keywords Zero-bit watermarking algorithm , Security, Attack, Subspace Estimation INTRODUCTIONIf watermarking robustness deals with the performance of a watermarking scheme against common processing operations (re-compression, transcoding, editing operations), the Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. security of a watermarking scheme is addressed whenever an adversary is part of the game and tries to remove the watermark. There are two important families of security attacks:• Sensitivity attacks [1] aim at removing the watermark by using the watermark detector as an oracle,

Better security levels for broken arrows

“…This confirms Patrick Bas' results. 4 However, for BA and AWC embedding techniques with the regulated parameters N v = 1024 and N c = 256, SCD norm is always close to 1, this shows that the OPAST algorithm cannot estimate the secret subspace any longer.…”

“…4 We sum up its main process here: 1) Through a huge number of observations, the fast and efficient OPAST algorithm estimates the projection matrix W, whose size is N s × N p , here N s is the number of the wavelet coefficients to be watermarked and N p represents the number of principal components. In order to assess the performance of his subspace estimation algorithm, he used the Square Chordal Distance (SCD) between the secret subspace span(S C ) and the estimated subspace span(Ŵ) during this step.…”

“…4 In this algorithm, all the bin leaders are updated at each iteration (step 4 to step 6). This operation does a lot of repetitive work, and wastes a lot of computing power.…”

Pit Stop for an Audio Steganography Algorithm

“…We could model the cover sample from other, but temporally close stego samples: A similar approach was used during the BOWS-2 contest to estimate the unmarked magnitude of wavelet coefficients from its surrounding [13]. This has been successful, because the piece of watermark in s i was independent of the watermark in the samples of the surrounding.…”

Towards robust and secure watermarking

“…The embedding is considerably modifying the power distribution of the signal in the secret subspace. With OPAST, an available implementation of the Principal Component Analysis (PCA), the authors of [3] succeeded to disclose the secret space of the original version of Broken Arrows, pointing out a lack of security. In a recent paper, we proposed a counter-measure to this attack [15].…”