Image process is a key process to the fish-like robot polo game. A good image process system is one of the basic elements of winning the game. But in the past games, it was before the game begun that we made effort to adjust the parameters to adapt to the environment. Obviously, this kind of method couldn't meet the reality of the changeable competition environment. This paper proposes a self adaptive image process method based on feedback and study. It adjusts parameters used in the image process progress by studying the intermediate results it produced. In this way, the parameters of the program can be maximally adaptive to the environment. Large number of experiments shows that this processing scheme can achieve good recognition results. It meets the requirement of the changeable fish-like robot polo game environment.