A viewpoint detection unit detects a user viewing a stereoscopic image, including a parallax image of a subject as viewed from a predetermined position defined as a reference view position, and tracks a viewpoint of the detected user. A motion parallax correction unit determines, if a speed of movement of the viewpoint becomes equal to or higher than a predetermined level, an amount of motion parallax correction for the parallax image, on the basis of an amount of movement of the viewpoint, so as to generate a stereoscopic image corrected for motion parallax, generates, if the speed of movement of the viewpoint subsequently becomes lower than a predetermined level, a stereoscopic image by changing the amount of motion parallax correction in steps until the parallax image return to parallax images as seen from the reference view position.