Abstract:Volume rendering, a technique for visualizing 3D arrays of sampled data, is computationally very expensive. Interactive volume rendering of large volumes (over 10243) is in demand for such applications as biomedical simulation and geoscience. Existing software optimizations can barely meet the challenge of today's increasingly large datasets. Even the fastest workstations equipped with special hardware are quickly overwhelmed. So far, none of the current special-purpose volume rendering hardware designs achieve real-time frame rate for large volumes at an acceptable hardware cost. In this dissertation, we propose an image-based framework which avails the coherence between neighboring frames during navigation of a volume dataset. The previously fully volume rendered view is cached as a keyview or keyframe. When generating the current novel view, instead of ray-casting the whole image, rays are cast only through those pixels which represent previously hidden objects. When the novel view camera moves only slightly away from that of keyview, the ray-cast pixels occupy only a small percent of the current view. Consequently, this gives us potential for vast speedup over traditional volume rendering. The rest of the novel view is generated directly from the keyview through inexpensive image transformation. To achieve this, we first construct a geometric iso-surface model from the volume, where the iso-value is deducted from the input transfer function. We further texture-map the keyview onto the constructed geometric model through projective texture mapping and then project it to the novel view. We have explored separate algorithms optimized for several stages of the framework: an algorithm for generating levels-of-detail representation of regular volumes—for efficient and levels-of-detail iso-surface extraction—and two algorithms for high quality and efficient texture mapping featuring backward and forward processing orders. To accelerate the volume rendering of keyview, we have developed a group of methods for decomposing 3D volume rotation into shear transformations which lend themselves to a feasible hardware or parallel implementation. Even though these algorithms are designed within this framework, we have demonstrated that they can be extended to serve other numerous general applications. We have applied our framework to several practical navigation systems. We have achieved an average of a magnitude speedup over that of traditional volume rendering and obtained an interactive frame rate on a case study of a virtual colon navigation of a patient's 5123 colon data. We have also extended the framework to interactive rendering of other data types, such as height field terrain and photograph sequence, and demonstrated the effectiveness of these extensions.

Hardware-Assisted Adaptive Volume Rendering for Mass Data

Real-Time Rendering of Highly Complex Dynamic Scenes Based on Parallel Multi-Core Architectures

Real-time volume rendering of large datasets on standard PC hardware

A Scalable Parallel Software Volume Rendering Algorithm for Large-Scale Unstructured Data

Parallel Volume Rendering Based on LOD Method

Adaptive voxels: interactive rendering of massive 3D models

Depth-Box VDB: Accelerate Sparse Volume Rendering with Depth Maps Through Voxel Database

Partition Based Real-Time Rendering Method for Large-Area Terrain Data

Accelerated rendering method based on area division

Real-time visualization of large volume datasets on standard PC hardware

View Dependent Layer Sampling: an Approach to Hardware Implementation of Volume Ray Casting

Animating and rendering one million trees in real time

High-quality multi-resolution volume rendering in medicine

Hardware Assisted Fast Volume Rendering with Boundary Enhancement

Depth-Based Feature Enhancement for Volume Visualization

Balancing Cpu And Gpu: Real-Time Visualization Of Large Scale 3d Scanning Models

An Efficient Out-of-core Volume Ray Casting Method for the Visualization of Large Medical Data Sets.

High Quality Real-Time Rendering of Large Scale Point Model

Realtime Ray Tracing on a Hibrid Parallel Architecture

A Hardware Assisted Ldi Building Algorithm With Application To Hybrid Volume Rendering

Image-based Volume Rendering