Non-Pinhole Approximations For Interactive Rendering

Depth images have been used to approximate scene geometry in a variety of interactive 3-D graphics applications. In previous work, images were constructed using orthographic or perspective projection which limits approximation quality to what is visible along a single view direction or from a single viewpoint. This paper shows that images constructed with non-pinhole cameras improve approximation quality at little additional cost provided that the non-pinhole camera offers fast projection. For such a camera, the fundamental operation of ray depth-image intersection proceeds efficiently by searching along the one-dimensional projection of the ray onto the image. In the context of two-camera configurations, our work extends epipolar geometry constraints to non-pinholes. We demonstrate the advantages of non-pinhole depth images in the context of reflections, refractions, relief texture mapping, and ambient occlusion.

Paul Rosen, Voicu S Popescu, Kyle Hayward, and Chris Wyman. Non-Pinhole Approximations For Interactive Rendering. IEEE Computer Graphics and Applications (CG&A), 2011.

@article{rosen2011non,
  title = {Non-Pinhole Approximations for Interactive Rendering},
  author = {Rosen, Paul and Popescu, Voicu S and Hayward, Kyle and Wyman, Chris},
  journal = {IEEE Computer Graphics and Applications (CG&A)},
  volume = {31},
  pages = {68--83},
  year = {2011},
  note = {Feature Article. 2011 Best Paper Award Runner-up.},
  abstract = {Depth images have been used to approximate scene geometry in a variety of
    interactive 3-D graphics applications. In previous work, images were constructed using
    orthographic or perspective projection which limits approximation quality to what is
    visible along a single view direction or from a single viewpoint. This paper shows that
    images constructed with non-pinhole cameras improve approximation quality at little
    additional cost provided that the non-pinhole camera offers fast projection. For such a
    camera, the fundamental operation of ray depth-image intersection proceeds efficiently
    by searching along the one-dimensional projection of the ray onto the image. In the
    context of two-camera configurations, our work extends epipolar geometry constraints to
    non-pinholes. We demonstrate the advantages of non-pinhole depth images in the context
    of reflections, refractions, relief texture mapping, and ambient occlusion.}
}