Background registration for foreground separation in 2D–3D video conversion

This paper presents a new method for stereoscopic 3D video generation from single 2D video. One of the stereo pair is taken as the 2D frame itself, and the other one is obtained by shifting it horizontally according to the disparity map. Disparity map is obtained from motion between frames and the geometrical structure of the scene. The disparity difference between foreground and background on the same horizontal rows of the frame is extracted via background registration based on the motion. Thus, the disparity of the foreground is separated automatically from the disparity of the background. After this procedure, the resulting frame is shifted according to relative depth information of the background structure where the shift values can be controlled by the user. This new frame and the original 2D frame are then used to build the 3D stereo frame.