This paper proposes an image watermarking scheme for copyright protection based on Independent Component Analysis (ICA). In the suggested scheme, we divide the original image into blocks and use the ICA to project the image into a basis with its components as statistically independent as possible. The data embedding is carried out in cumulative form in the ICA basis and also the ICA is used for watermark extraction. The extraction process could be either non-blind, through a straightforward procedure, or blind, via a tricky method we propose. Experimental results show that the new method outperforms some well-known image watermarking methods in spatial or frequency domain and is robust against various attacks, including noise addition, resizing, low pass filtering, multiple marks embedding, JPEG compression, gray-scale reduction, and cropping parts of image.