The crucial of design a robust image watermarking scheme is cope with geometrical attacks in transform domain.In this paper, a novel feature-based image watermarking scheme resilient to the geometrical attacks is proposed. Firstly, the harris-affine detector is utilized to extract invariant feature points from the host image. Then, by the feature scale of the local structure, the feature regions are adaptively computed and normalized. Finally, several copies of the digital watermark are embedded into the nonoverlapped local feature regions by quantizing the magnitude vectors of the DFT middle frequency. During detection, the resynchronization is achieved through feature points. And the embedded sequence can be extracted by “majority principles” without resorting to the original host image. Experimental results show that the proposed algorithm has good transparency and is very robust to common image processing and geometric attacks.