Screen Space Reflection Techniques

Abstract

Ray tracing is a rendering technique in computer graphics that can simulate a variety of optical effects, such as reflection from smooth surfaces, refraction through transparent objects, and light scattering on rough surfaces. Ray tracing can produce visual realism of a higher quality than other rendering techniques, such as rasterization techniques, but at a much higher computational cost. Screen Space Reflection (SSR) is a group of approximation techniques that utilize data already generated by common rasterization techniques, such as deferred shading, to produce limited reflection effects.

Most rasterization algorithms use two basic data structures, an image buffer storing the colour of a surface point visible from the camera at each pixel, and a depth buffer (or Z-Buffer) storing the depth from the camera to the corresponding surface point. For each surface point, SSR techniques generate a reflection ray and project the reflection ray onto the depth buffer. Values in the depth buffer along the projected path are compared with the depth of the reflection ray to determine a potential intersection. A traversal process along the projected path is necessary. If an intersection is found, the corresponding colour in the image buffer contributes to the reflection. Several SSR techniques exists. In this research, SSR techniques are defined using a common algorithm schema and five noteworthy SSR techniques including ray marching, digital differential analyzers (both conservative and non-conservative), and hierarchical depth buffers (both minimum and minimum-maximum) are implemented as instances of this schema for the purpose of performance analysis.

In the performance analysis, the average GPU time for traversing a projected path in the depth buffer was recorded for each of the five SSR techniques on different testing scenes, different image resolutions, and under variable control parameters associated with each technique. The analysis shows a statistically significant difference in average traversal time between different SSR techniques for 98% of the test configurations. Visualizations are also generated to facilitate the analysis. Detailed analysis results are presented in the thesis