Superview 3D image warping for visibility gap errors

Abstract

3D image warping is an image-based rendering technique that allows nearby views of a 3D environment to be efficiently extrapolated from a single 2D reference image that includes depth information. This method allows the real-time rendering of a virtual environment on low powered consumer devices such as PDAs, cellular phones, e-Books, etc. However one major drawback with this method is the frequent occurrence of visibility gap errors caused by a limited field of view. Visibility gaps are regions for which the new viewpoint has no information and often occur during a viewpoint rotation or lateral translation. The resulting warped image contains significant areas of unsightly holes. This paper proposes using a reference image larger than the display size with a greater field of view (superview), to reduce or eliminate visibility gap errors. However warping a larger reference image increases rendering time. In order to address the problem of the reduced frame-rate, acceleration methods such as image sub-sampling, pixel averaging and clipping are presented. It is concluded that use of an oversize reference image in conjunction with sub-sampling and clipping of the reference image produces a better quality warped image for a given frame-rate.