Abstract: A robust image watermarking in affine covariant regions is proposed. First, the affine-invariant point detector is used to extract feature points of a given host image. A set of stable and non-overlapped affine covariant regions are then chosen via selection criterion based on the minimum spanning tree clustering algorithm. Secondly, image normalization and dominant gradient orientation alignment are performed on these feature regions to achieve scaling and rotation invariance. Finally, watermarks are directly embedded in these feature regions. Extensive experimental results demonstrate that the proposed scheme is robust against common image processing operations, geometric attacks, and combined attacks.