Rotation Angle Detection Using a Pilot Signal from Rotated Stego-Image

Robust watermarking methods are necessary to resist attacks. With conventional watermarking methods, watermarks are extracted and their errors are corrected regardless of the type of attack. For example, error-correcting codes and spread spectrum are used to correct watermark errors. To reduce watermark errors, embedding methods in the frequency domain, such as DCT and DFT, and the use of scale invariant feature transform have been proposed. These methods are effective for non-geometric attacks such as additive noise and distortion. However, they do not work well against geometric attacks due to the lack of synchronization. It is necessary to perform optimal synchronization for geometric attacks such as scaling, rotation, cropping, etc. We focus on the rotation attack and propose a method of estimating the rotation angle using a grid-shaped pilot signal. While suppressing the original image-derived slant, the slant of the pilot signal can be emphasized and detected. Thus, the rotation angle of a stego-image can be accurately estimated.