Layered Depth Images

A method that does not need special knowledge of the model, and which is not only applicable to trees or plants, is presented by Shade et al. [195, 196]. They solve the problem of the earlier-mentioned occluded pixels and resulting holes by storing all occluded pixels during the production of a source image for the
image-based description. To yield a satisfactory result, a raytracing algorithm is applied and, thus, the method can only be applied to computer-generated models.

Подпись: extended raytracingStarting from the center, rays are directed through each surface of a unit cube, which surrounds the center. Each of the six surfaces is used just like the projec­tion surface for obtaining a conventional raytracing image, and has accordingly a number of pixels that are penetrated by the rays from the center. However, along these rays not only is the first intersection point with the geometry stored, but also all other intersections with the scene, including their respective depths. This results in a so-called layered depth image, or LDI. Here, each ray consists of a list of intersection points with color, transparency, and depth.

Figure 10.6

Scene from Fig. 8.18, illustrated using layered depth images. Good image quality in mid-distance; images at a closer distance show strong pixel effects and have to be replaced with polygonal representations


Layered Depth Images

This information can be used to compute the image information for a new cam­era angle through reprojection. Since we are dealing here with many points that now have to be processed, Shade et al. [195] develop an optimized calculation. Figure 10.6 shows a plant scene rendered with LDI. For a medium distance, acceptable results are yielded, while with closer distances a polygonal approx­imation still has to be applied.