Depth of Field: Point Splatting on Per-Pixel Layers

Christoph Kollmann
Depth of Field: Point Splatting on Per-Pixel Layers
[thesis]

Information

Abstract

This bachelor’s thesis is about a depth of field rendering algorithm using point splatting on per-pixel layers. The intuitive procedure just uses specific adjustable parameters for the lens of a camera to generate the desired effect in real-time. Using nothing but the depth information of a scene, the depth of field can be computed. The so called circles of confusion are the representing value of the blur rate for objects that are out of focus. Depending on the distance between the focal plane and the current pixel, an expansion takes place because of the redirecting nature of a lens. The following required information is retrieved using an intensity distribution function. To avoid intensity leakage and provide a full opaque picture, the entire effect is done discrete on the screen space. This prevents subsampling and enables almost correct alpha value normalization. As stated before, an intensity distribution function is required, but in a discrete form. This is an addition to the paper, where that function is preprocessed in the application and provided for the shader directly. A high resolution texture of the desired intensity distribution function is provided as input to a shader program that renders multiple discrete versions. Another further addition prepares the texture by normalization, consuming more memory, but saving lots of time during computation. The last part of this paper includes a defocused highlights implementation, where a tone-mapping is applied to mimic them. In general, the luminance of a pixel is computed by a simple formula, summing up the three colors with adjustment factors. Depending on this value and some parameters, the color is adjusted where bright regions are enhanced to construct this effect. This thesis implements and extends this work and is used to test its performance after seven years when it was published. Using different configurations for the effect, the provided stable framerates of the results are used as metric for benchmarking.

Additional Files and Images

Additional images and videos

image: image image: image

Additional files

thesis: thesis thesis: thesis

Weblinks

No further information available.

BibTeX

@bachelorsthesis{Kollmann-2015-DoF,
  title =      "Depth of Field: Point Splatting on Per-Pixel Layers",
  author =     "Christoph Kollmann",
  year =       "2015",
  abstract =   "This bachelor’s thesis is about a depth of field rendering
               algorithm using point splatting on per-pixel layers. The
               intuitive procedure just uses specific adjustable parameters
               for the lens of a camera to generate the desired effect in
               real-time. Using nothing but the depth information of a
               scene, the depth of field can be computed. The so called
               circles of confusion are the representing value of the blur
               rate for objects that are out of focus. Depending on the
               distance between the focal plane and the current pixel, an
               expansion takes place because of the redirecting nature of a
               lens. The following required information is retrieved using
               an intensity distribution function. To avoid intensity
               leakage and provide a full opaque picture, the entire effect
               is done discrete on the screen space. This prevents
               subsampling and enables almost correct alpha value
               normalization. As stated before, an intensity distribution
               function is required, but in a discrete form. This is an
               addition to the paper, where that function is preprocessed
               in the application and provided for the shader directly. A
               high resolution texture of the desired intensity
               distribution function is provided as input to a shader
               program that renders multiple discrete versions. Another
               further addition prepares the texture by normalization,
               consuming more memory, but saving lots of time during
               computation. The last part of this paper includes a
               defocused highlights implementation, where a tone-mapping is
               applied to mimic them. In general, the luminance of a pixel
               is computed by a simple formula, summing up the three colors
               with adjustment factors. Depending on this value and some
               parameters, the color is adjusted where bright regions are
               enhanced to construct this effect. This thesis implements
               and extends this work and is used to test its performance
               after seven years when it was published. Using different
               configurations for the effect, the provided stable
               framerates of the results are used as metric for
               benchmarking. ",
  month =      sep,
  type =       "Bachelor Thesis",
  address =    "Favoritenstrasse 9-11/186, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology",
  keywords =   "depth-of-field, deferred rendering, real-time rendering",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2015/Kollmann-2015-DoF/",
}