A robust color image watermarking scheme in the fusion domain based on LU factorization

Since most of the current watermarking schemes neglect the need of real time and resisting malicious geometric attacks, this paper proposes a robust color image watermarking scheme based on the deep fusion theory between spatial domain and LU factorization to overcome these problems. The color watermark image is embedded to the color host image by modifying high correlation elements of matrix L, and an algorithm derived for the first time in this paper enables this process to be completed directly in the spatial domain without the real frequency domain conversion, which simplifies the operation process and reduces the computational complexity. Furthermore, an effective image correction algorithm is innovatively proposed, which detects and locates image vertices through Hough transform with strong anti-interference ability, corrects any quadrangle through perspective transform, and finally performs anti-aliasing processing on image edges based on the idea of fast approximate anti-aliasing algorithm. The significant contributions are summarized as: 1) the elements of LU factorization can be calculated directly in the spatial domain; 2) the geometrically attacked watermarked image can be corrected; 3) the fast approximate anti-aliasing is used to resist geometric attack. The experimental performances are shown as follows: 1) the average values of PSNR and SSIM are above 40 dB and 0.96 respectively, which indicates the invisibility of the proposed scheme using variable quantization steps is enough to meet the requirement of good image quality; 2) the average values of NC and BER are above 0.93 and about 0.1 respectively under different attacks, which emphasizes the proposed image correction algorithm can effectively deal with images under some geometric attacks with strong robustness; 3) the total time is about 0.14 s, which highlights the proposed fusion-domain algorithm can reduce nearly 63 % of the running time with high real-time feature.