On the luminance overflow in spread spectrum robust image watermarking schemes

In spread spectrum robust image watermarking schemes, images are watermarked in frequency domains and transformed back to spatial domains. Different from unlimited frequency coefficients, the volumes in spatial domains are usually limited, therefore luminance overflow frequently occurs on some pixels in watermarked images especially for those with high average luminance. In all existing solutions to this problem, noticeable decreases are generally introduced into the results of watermark detections, and the robustness of watermarking schemes is tampered severely as a result. However, since any operations in frequency domains would be diffused to all pixels, it is hard to investigate the effect of luminance overflow on watermark detections in frequency domains. The authors propose an approximating calculation of detection values in spatial domains for spread spectrum watermarking schemes. With the proposed approximating method, it is shown that the detection values are significantly decreased for images with higher average luminance and fewer textures. Finally a new self-adaptable solution is established to adjust the luminance of watermarked images in a block-wise style, where two adjusting factors are calculated to obtain the best fidelity and robustness. Experiments and analysis show that the new solution is effective improving the fidelity and robustness of object watermarking schemes.