Invariant Image Watermarking Based on Local Feature Regions

In this paper, a robust image watermarking approach is presented based on image local invariant features. The affine invariant point detector is used to extract feature regions of the given host image. Image normalization and dominant gradient orientation alignment are performed to these feature regions. Thereafter, circular watermark pattern are embedded in these image normalized patches. After the above procedures, full affine invariance can be achieved over these normalized patches. A large amount of experiments demonstrate that the proposed approach is robust against different attacks, such as common image processing, rotation, scaling, cropping, and several other geometric distortions.