Rendering Trees from Precomputed Z-Buffer Views

Abstract

Abstract. Parallel projection z-buffer images are precomputed for a number of preset viewing directions on the unit sphere. Using the depth information, we can reconstruct a 3-D point from each image pixel. Then parallel or perspective views can be found from any other viewpoint by rotating the 3-D points from the nearest three or four precomputed views into the proper position and projecting them onto an output z-buffer. A temporary z-buffer image, constructed in the same way for a viewpoint at the light source, is used for a z-buffer shadow algorithm. The rendering algorithm is applied to broad leaf and needle leaf trees, and tested in animation. Our rendering algorithm is similar to that in Chen and Williams [1], with differences listed below.

Nailboards: A Rendering Primitive for Image Caching in Dynamic Scenes

“…Re-projecting pixels, view interpolation, rendering trees from pre-computed zbuffer views [9] and RGB∆ textures already demonstrate four such trade-offs but others may be possible.…”

Hierarchical Image-Based Rendering using Texture Mapping Hardware

“…However this does not completely eliminate gaps from surfaces which are at too steep a slant in the precomputed view, and cannot show surfaces that are occluded in the precomputed image. Both Schaufler [5,6] and Max [4,7] reproject several views to help solve these problems. In this paper we use all six views along the positive and negative directions on the x, y, and z axes.…”

“…Shade et al [3] and Max [4,7] used multiple depth layers in the precomputed images, whose depths were stored individually in a sorted list at each pixel, and thus could not be easily generated in standard z-buffer hardware. To take advantage of hardware rendering, we slice the object by hardware z clipping into multiple slabs, and reproject each of them by the method of Schaufler [5].…”

“…McMillan and Bishop [2] showed that a single image can be processed in a "painter's" order that does not require a z-buffer for the output image, and Shade et al [3] extended this to multiple z-layers at each pixel. Max [4] and Shade et al [3] used the multiple layers to include more of the depth complexity in the pre-computed images, which might become "disoccluded" in the new view.…”

Image warping for compressing and spatially organizing a dense collection of images

“…In recent years, image-based modeling and rendering approaches (IBMR) have addressed this problem [1]. They do so by directly resampling images to generate new views, thus avoiding the need for a detailed geometric model.…”