In this paper a blind image watermarking that can resist to rotation, scaling and translation (RST) attacks is proposed. Based on the spread spectrum, the watermark is modulated before embedding. The log-polar mapped discrete fourier transform (LPM-DFT) magnitude of a disk, a part of the origin image, constitutes the RST-invariant domain, where the origin of the LPM is the center of the disk and the sampling rates of the LPM are constant. After the middle frequency band of LPM-DFT magnitude, namely the watermark-embedding domain, is grouped according to the watermark length, the watermark is embedded by adjusting the difference between the two sub-region energy in each group. To improve the imperceptibility, the watermark-embedding domain is shuffled before embedding and the watermark is not embedded directly into the watermark-embedding domain. In watermark detection procedure, neither the original image nor any knowledge about the distortions is required. Experimental results show that the proposed scheme is very robust against RST distortion and common image processing attacks.