The watermarking method has emerged as an important tool for content tracing, authentication, and data hiding in multimedia applications. We propose a watermarking strategy in which the watermark of a host is selected from the robust features of the estimated forged images of the host. We introduce the concept of watermark space which is obtained from Monte Carlo simulations of potential pirate attacks on the host image. The subspace approach can be used for symmetric and asymmetric watermarking scheme for copyright protection. Both of them reveal the phenomena of a high detection probability and low false positive probability. In either scheme, significant efforts have been made to explore the probability of successful hacking into our system. The result shows our schemes sustain most known attacks.