Lightfield Rendering Based on Depths from Physically-Based Volume Rendering
United States Patent Applications 20180227568
The present embodiments relate to physically-based volume rendering. In a medical imaging example, an image from a three-dimensional (3D) scan of a patient is rendered for visualization. Physically-based visualization techniques produce global illumination effects in computer-generated graphics that mimic the real-world interaction of light with various 3D objects or tissues. Monte Carlo path tracing is a useful rendering technique for such light transport computations, where the natural light phenomena are modeled using a stochastic process. The physically-based volume rendering results in physically plausible images that are often easier for the human brain to interpret when compared to the more analytical images from traditional volume rendering, such as ray casting or direct volume rendering. However, evaluation of the rendering integral in physically-based volume rendering may require thousands of stochastic samples per pixel to produce an acceptably noise-free image. Depending on the rendering parameters and implementing processor, producing one photorealistic image with physically-based volume rendering may be on the order of seconds for interactive workflows and multiple hours for production-quality images. Light field rendering allows much faster generation of photorealistic images, but depends on a large amount of input data that has to be produced in an expensive preprocessing step. This preprocessing step shifts the time demand form the phase of the user interaction to the preprocessing step. This also allows fast rendering on low-powered devices, like tablets or cell phones, while shifting the computationally expensive task of preprocessing to dedicated computer servers.