Flaws in the warped image

Figure 7.21: If the viewing position changes, then a visibility event might be encountered. This means that part of the object might suddenly become visible from the new perspective. In this sample, a horizontal shift in the viewpoint reveals a side of the cube that was originally hidden. Furthermore, the top of the cube changes its shape.

Image warping due to orientation changes produces a correct image in the sense that it should be exactly what would have been rendered from scratch for that orientation (without taking aliasing issues into account). However, positional changes are incorrect! Perturbations in $x$ and $y$ do not account for motion parallax (recall from Section 6.1), which would require knowing the depths of the objects. Changes in $z$ produce similarly incorrect images because nearby objects should expand or contract by a larger amount than further ones. To make matters worse, changes in viewpoint position might lead to a visibility event, in which part of an object may become visible only in the new viewpoint; see Figure 7.21. Data structures such as an aspect graph [254] and visibility complex [256] are designed to maintain such events, but are usually not included in the rendering process. As latencies become shorter and prediction becomes better, the amount of perturbation is reduced. Careful perceptual studies are needed to evaluate conditions under which image warping errors are perceptible or cause discomfort. An alternative to image warping is to use parallel processing to sample several future viewpoints and render images for all of them. The most correct image can then be selected, to greatly reduce the image warping artifacts.