Abstract: Under the assumptions that gray-scale watermark data may be summarized as a known, zero-mean, binary valued sequence and the original image data as an independently and identically distributed random sequence from a Weibull distribution, we show that this embedding strength should be estimated through simple computations based on the average gray-scales of both the watermarked image and original image. An approach based on moment estimation is presented. Experimental results show that this approach is effective for De Rosa et.al.'s watermark detector. Moreover, according to the embedding rules of addition, multiplication and fusion, we also compared and analyzed their performance using PSNR. Conclusions obtained from experiments are useful to the further research.