Approximate Ray-Casting Volume Rendering Based on Adaptive Sampling

Real-time high-quality volume rendering in Virtual Reality (VR) and Head-Mounted Displays (HMDs) presents significant computational challenges due to their high resolution and framerate demands. While adaptive sampling methods have been developed to address performance issues in traditional RayCasting, they often rely on inefficient trilinear interpolation and lack global error control, which limits rendering speed and quality respectively. This paper introduces a novel approximate Ray-Casting framework that addresses these limitations. First, based on an analysis of commonly used biomedical volumes in the visualization community, we perform efficient actual-sampling at voxel boundaries by 2D-texture-based bilinear interpolation, while internal virtual-sampling points are derived via linear interpolation. Second, we introduce a global error control mechanism that dynamically allocates virtual-samples based on local error contribution, ensuring bounded rendering error. Evaluations on 5 biomedical volumes demonstrate our method achieves superior rendering quality at 90 frames per second (fps) on an HMD, outperforming state-of-the-art adaptive sampling techniques in both quality and speedup. A user study with 30 participants confirms the perceptual superiority of our approach.