Description

Depth-of-field effect is widely used in computer graphics to add an additional degree of realism in the synthetically generated scene. The traditional approach involves image-space filtering, but things are getting complicated when one focuses on a highly reflecting object that reflects an environment which is far away. A straightforward approach fails to work fast because the complexity of filtering grows as a square with a filter kernel size. The goal of this project is to develop an approach for fast correct pre-filtering of cube maps of arbitrary size. On of the important applications of this approach is photorealistic gemstone rendering, where environment is reflected and refracted in a gemstone.

Task

This project will include the following subtasks:

Basic implementation

The application should produce correct filtering environmental map for given parameters (focus distance, distance to environment, aperture etc). It should produce correct results, e. g. be continuous on the borders

GPU/multicore accelerated implementation

There are several ways how to accelerate this techniques on GPU/other parallel architecture. One of them should be implemented in this subtask

Image processing acceleration

Circle of confusion is basically a constant-value filter. For such a filter, an acceleration can be performed which reduces computation complexity from O(n * n) down to O(n)

Requirements

Good programming skills, some experience with shader programming, good knowledge and interest in computer graphics. Besides, interest in image processing is welcome.

Environment

The student is free to select any environment and programming system he likes. There is only one limitation - the source code should compile and execute in MS VS 2008 environment.

Additional Information

Contact: peter.sikachev {ät} cg . tuwien . ac . at
Please, contact me in English, as I usually skip German e-mails.

• Computer Graphics Group
• Courses
• Projekte
• Environment Map Filtering