Rotation,scaling and translation(RST) attacks can desynchronize the watermark detection so that many watermark systems fail.In this paper,a RST invariant watermark is proposed by using local Zernike moments.As the Zernike magnitude moments of a normalized image are RST invariant,the watermark image is embedded into the local area of the original image after reconstruction.And the Zernike magnitudes of the area are extracted as the eventual watermark vector.In detection,the Zernike vector of the selected area is first extracted,and the root-mean-square-error(RMSE) is employed to decide the presence of the watermark.The local Zernike moments makes the watermark more robust to rotation attacks than the global moments.The watermark is also robust to scaling as well as regular attacks such as JPEG compression and added noises.