We proposed the three dimensional short-term memory image related to human memory and human vision. Three dimensional short-term memory image composed of the three dimensional features reconstructed from the binocular foveated vision. The next fixation point is determined statistically from the feature points in the image projected from the 3-D short-term memory image. We show the effectiveness by simulating the binocular eye movements for the virtual world generated computer graphics and for the real world using pan-tilt cameras. It is found that the viewing world does not change suddenly though the eye moves quickly by introducing the 3-D short-term memory.