In this paper we proposed a novel technique for watermarking facial images into different background images in an invisible manner. This facial image can be extracted blindly from the watermarked image without any prior knowledge about the background image (blindly). More over the proposed technique can still deliver an extracted face image even in the presence of noise, compression or cropping of the watermarked image, but with some visual quality degradation. This embedding/extraction methodology is fake-free and robust against compression, noise and cropping attacks. The proposed approach consists of an insertion part and an extraction part. The proposed approach is based upon the natural preserve transform technique that achieves a balance between energy concentration and energy distribution; hence it is best used for watermarking images. Extensive simulation results on several facial and background images, that demonstrate the effectiveness of the proposed technique, are illustrated in comparison with current watermarking methods in the literature.