Statistical Attack Resilient Data Hiding

Abstract

In this paper, we propose a discrete cosine transform (DCT) based spread spectrum data-hiding algorithm that is resilient to statistical attacks. Unlike other spread spectrum based data-hiding algorithms, the proposed algorithm does not introduce a low-pass filtering effect in the histogram of the stego image. The distance between the center of gravity (CoG) as defined by [14] of the unmarked host and the stego images was reduced by 74% in the proposed algorithm. The proposed algorithm is also resilient against the Chi-Square attack and does not compromise on robustness or capacity to achieve this goal. When compared to the generic block based DCT data-hiding scheme, the proposed algorithm provides a 41% reduction in the relative entropy between the host and stego images. In other words, the proposed algorithm is 41% more secure than generic DCT based data-hiding algorithms when measured in terms of relative entropy. The proposed algorithm also provides statistical resilience against a steganalysis attack specifically designed for block DCT datahiding algorithms [29]. The proposed algorithm is robust against a variety of image manipulating attacks such as noise addition, filtering, blurring, sharpening, JPEG compression etc. In the cases of dislocating attacks such as blurring and despeckling, the bit error ratio (BER) was 0.1045 and 0.0435 respectively, thereby yielding retrieval rates of over 89% and 95% respectively. In the case of low quality JPEG attack (Q-30) the retrieval rate was 92%. In the case of noise addition attacks, the retrieval rates were more than 92%.