Error Metrics for Smart Image Refinement

Abstract

Scanline rasterization is still the dominating approach in real-time rendering. For performance reasons, realtime ray tracing is only used in special applications. However, ray tracing computes better shadows, reflections, refractions, depth-of-field and various other visual effects, which are hard to achieve with a scanline rasterizer. A hybrid rendering approach benefits from the high performance of a rasterizer and the quality of a ray tracer. In this work, a GPU-based hybrid rasterization and ray tracing system that supports reflections, depth-of-field and shadows is introduced. The system estimates the quality improvement that a ray tracer could achieve in comparison to a rasterization based approach. Afterwards, regions of the rasterized image with a high estimated quality improvement index are refined by ray tracing.