Document Type
Report
Abstract
Image-based rendering (IBR) techniques have the potential of alleviating some of the bottlenecks of traditional geometry-based rendering such as modeling difficulty and prohibitive cost of photorealism. One of the most appealing IBR approaches uses images enhanced with per-pixel depth and creates new views by 3D warping (IBRW). Modeling a scene with depth images lets one automatically capture intricate details, which are hard to model conventionally. Also, rendering from such representations has the potential of being efficient since it seems that the number of samples that need to be warped is independent of the scene complexity and is just a fraction above the number of samples in the final image. However, selecting the subset of reference-image samples that need to be warped to generate the new view is a very difficult task. We present the vacuum buffer algorithm, and its use within a sample-selection method. Like other techniques, our method proceeds by considering samples of reference images that were acquired from locations close to the current camera position. Unlike other techniques however, our method offers a conservative estimate on whether samples of visible surfaces were potentially missed and it also points to the scene locations where such surfaces might be. The vacuum buffer is essentially a generalized z-buffer and it measures what sub-volumes of the current view-frustum are yet to be determined. Another impo1tant difference is that our method uses the current view, which allows it to reduce the number of chosen samples more than other methods that offer a sample selection solution to be used for several desired camera views. The tradeoff for using the current view is having to solve the sample selection at each frame. By exploiting the coherence in the reference images, groups of nearby samples become the actual primitive, which massively reduces the total cost.
Publication Date
11-5-2001
Recommended Citation
Popescu, Voicu, "The Vacuum Buffer" (2001). Link Foundation Modeling, Simulation and Training Fellowship Reports. 38.
https://repository.fit.edu/link_modeling/38
Standard cover form for report
Comments
Link Foundation Fellowship for the years 2000-2001.