Image watermarking can be distinguished between methods that hide information by adding and multiplying. In some applications, not the more information embedding is, the more performance is especially for multi copyright protection. In this paper, robustness performance of these two watermarking methods are compared with the same conditions, including original image, watermark signal, fidelity, detection requirement such as detection probability or/and false alarm probability. Watermark capacity is computed according to the requirements of watermarking real applications with the same conditions. How to compute capacity is described in detail. Independent Component Analysis (ICA) is utilized to extract the watermark blindly. Experimental results have demonstrated that the proposed watermarking technique with optimum capacity can meet requirements of different real watermarking applications. It is indicated that this new watermarking technique has superior advantages over the existing ones in many aspects according to the real watermarking applications requirements. Experimental results also show that for many Stirmark attacks multiplicative watermarking has a better robustness than additive watermarking.