CMPT 461 Final Project Sunny Tam - PowerPoint PPT Presentation

1 / 14
About This Presentation
Title:

CMPT 461 Final Project Sunny Tam

Description:

Camera often produces a light ringing or 'donut' effect. Photographers prefer bokeh with softer edges. Lack of papers and formulas. Goals ... – PowerPoint PPT presentation

Number of Views:81
Avg rating:3.0/5.0
Slides: 15
Provided by: sun653
Category:

less

Transcript and Presenter's Notes

Title: CMPT 461 Final Project Sunny Tam


1
CMPT 461 Final ProjectSunny Tam
2
Background and Motivation
  • Bokeh Japanese term for anesthetic quality of
    out of focus blur
  • Different lens produce different types and shapes
    of blur

3
Background and motivation
  • PBRT provides basic out of focus blurring
  • Circular and uniformly sampled
  • Often unrealistic and undesirable
  • Camera often produces a light ringing or donut
    effect
  • Photographers prefer bokeh with softer edges
  • Lack of papers and formulas

4
Goals
  • To provide the ability to specify the shape and
    the distribution of the out of focus blurring
    effect
  • Implement in a easy to use, and readily
    expandable manner

5
PBRT depth of field
  • In essence a disk sampling problem

6
1D Non-uniform distribution
  • Along the radius pointed by a given theta

7
1D Non-uniform distribution
  • Exponential
  • -ln(u1)/a
  • Gaussian
  • ยต ssqrt(2)erfinv(2u1 1)
  • Triangular
  • a sqrt(u1(m-a)(b-a)) if u1 lt m
  • b sqrt((1-u1)(b-m)(b-a)) otherwise

8
Non-circular disk sampling
  • Specify a surface which is a subset of the unit
    disk

9
Non-circular disk sampling
  • From here any 1D distribution can be mapped along
    the length of h

10
Results
  • Implemented as part of the perspective camera
    class
  • Example of usage
  • Detailed instructions on website

Camera "perspective" "float fov" 50 "float
lensradius" 0.15 "float focaldistance" 2
"integer mode" 3 "integer shape" 3 "integer
power" 3
11
Results
  • Computationally expensive to generate high
    quality images

12
Challenges
  • Difficult to implement equations for the inverse
    cdf of a gaussian distribution
  • Requires error function calculation
  • Third party methods
  • Triangular disk difficult to sample correctly
  • Large difference in length of r
  • Sqrt the distribution of angles solves the
    problem conceptually but still appear to be
    imperfect

13
Conclusion
  • Need to accurately model real life camera
    behavior
  • Created a plugin that can readily accept more 1D
    distributions
  • Lower order polygons not realistic, but provide
    an artistic option

14
References
  • http//home.online.no/pjacklam/notes/invnorm/impl
    /natarajan/normsinv.h
  • http//en.wikipedia.org/wiki/Aperture
  • http//en.wikipedia.org/wiki/Bokeh
  • http//portal.acm.org/citation.cfm?id282449coll
    GUIDEdlGUIDECFID4354792CFTOKEN43069400
  • http//www.flarg.com/bokeh.html
  • Matt Pharr and Greg Humphreys, Physically based
    rendering
  • http//fendinadia.com/2007/11/modify-your-bokeh.ht
    ml
  • http//www.worldscibooks.com/mathematics/etextbook
    /5720/5720_chap1.pdf
Write a Comment
User Comments (0)
About PowerShow.com