1. Image Correction for Immersive Visualization Systems Alex Timchenko 02/03/2006

2. Project Background Immersive Visualization System TAMU TITF Grant 2002 - 2005 NSF Grant - Ongoing Principle Investigators Dr. Frederic Parke Dr. Donald House

3. Project Purpose Development of a modular low-cost immersive environment Visuals consists of a set of projectors back- projecting images on polygonal facets Have to utilize readily available technology and materials Powerful programming API to facilitate development of new applications for the system

4. 5-Facet Display Example

5. Need for Coherency Temporal coherency Each display must project a part of the world in-sync with the other projectors Achieved through network protocols Spatial coherency Two adjacent facets must display adjacent information with as little of a seam as possible

6. The Problem Image alignment on individual projectors We Want…We Get…

7. Problems with Projectors The projector is off-axis; lower distance shorter than upper distance Results in strong vertical perspective distortion

8. Problems with Projectors Perspective distortion Vertical distortion Keystone Horizontal distortion Keystone correction not as common More expensive Keystoning reduces image quality Difficult to manually align projector

9. Problems with Projectors Barrel Distortion Very difficult to manually correct Makes perfect alignment virtually impossible

10. Problems with Projectors Other Problems Non-uniform Intensity Non-uniform Color Color difference between projectors

11. Solutions Manual/Mechanical Solution Very difficult to precisely align No way to correct barrel distortion Software based solution Apply a real-time warp to the projected image, allowing the user to control the corners and midpoints of the polygon

12. Solutions First approach: Quad-Grid Each edge vertex is moveable Moving the corners corrects perspective distortion Moving the edge points corrects barrel distortion

13. Solutions First approach: Quad-Grid Good visual results Difficult to use Limited to quads Rendered with OpenGL evaluators

14. Solutions

15. Next Approach: Web Improved intuitive interface Applicable to n-sided facets Very simple structure

16. Solutions Next Approach: Web Difficult to find centroid Texture “swims” across seams Localized texture effect Creates slight shear and pinch No side-effects Unaffected area pinch

17. Solutions Next Approach: N-Grid Places a grid over an n-sided facet Distributes texture difference over a larger area Focuses distortion to corners Structure is difficult to handle OpenGL rendering issues

18. Solutions Other approaches? Spring-based meshes Treat interior points as particles Treat edges as springs Spring Stability Return to initial state

19. GPU Rendering Instead of relying on OpenGL default texturing, control the warping through the GPU Create a 2D displacement texture Access the displacement texture to get an offset, then access the image with the UV coordinates and the offset

20. GPU Rendering If we can find a displacement map for a mesh, we can use it Requires fairly advanced GPU Hardware Texture Access Texture Access Common Functionality Parallel Texture Access Texture Access Texture Access Advanced Functionality Sequential Texture Access

21. GPU Rendering Other advantages Color correction Easy to hue/color shift texel values Brightness correction Easy to adjust the brightness of texels Intensity falloff correction by altering brightness based on some grayscale image

22. Conclusion Image correction critical to immersive display systems Commodity projectors introduce problems Perspective and barrel distortion are difficult to correct manually Software solutions based on grid structures and real-time image warping produces best results

23. Thank you for listening!