In this paper, a robust algorithm with large data payload and high computational efficiency is proposed, which is suitable for real-time watermarking of JPEG or MPEG streams because it operates directly on DCT (discrete cosine transform) blocks. The proposed method is based on modifying the low-mid frequency DCT coefficients imperceptibly to modulate block energy. During the modulation, a theorem deduced from the Watson’s perceptual model is employed to restrict the modified magnitude of coefficients. This system is capable of embedding 2048 bits of information in images with dimensions 512×512 pixels. Experimental results indicate that the presented scheme is transparent and robust to significant valumetric distortions (including additive noise, low-pass filtering, lossy compression and valumetric scaling) and a part of geometric distortions.