An optimized robust and invisible digital image watermarking scheme in Contourlet domain for protecting rightful ownership

Imperceptibility, robustness, security and capacity are the main features of a digital image-watermarking scheme, and their simultaneous acquisition is a challenging problem. In the recently presented watermarking algorithms, the focus was mainly on increasing robustness against different attacks and the lack of a comprehensive and robust method to these attacks while maintaining acceptable imperceptibility and capacity is evident. By considering the watermarking as a multi-objective problem, this paper presents a new intelligent hybrid method to increase the robustness to a large number of attacks while increasing imperceptibility and embedding capacity. In the proposed algorithm, the cover image is decomposed into the set of non-overlapping blockes and by analyzing the eadge map of each block, the proper blocks are selected and transferred to the frequency domain using Contourlet Transform (CNT). Using the outstanding features of the CNT, i.e. multidirectional and multiresolution, the suitable frequency coefficients are selected which form the feature matrix for the embedding procedure. Then, SVD transform is performed on the feature matrix as well as the watermark image. In addition, to increase the robustness without losing the transparency, the dynamic scaling factors are automatically selected by PSO based on the attack test results, instead of using fixed or manually set the scaling factors. Moreover, to increase the security and solve the False Positive Problem (FPP), which is a common problem in the most SVD-based watermarking techniques, a new method is proposed to increase the partial information of matrix U to the singular values with a security key. Finally, the watermark image is inserted into the cover image based on the obtained optimal values of the scaling factors. The experimental and comparative results in terms of imperceptibility, robustness, capacity and security, demonstrate that our algorithm offers very good imperceptibility with a PSNR value of 57.31 dB and a significant enhancement in robustness against a wide range of geometric and destructive attacks. The proposed scheme also obtains a large capacity and sufficient security. Moreover, the proposed method is free of false positive detection error.