RAYS (Render As You See)

1
RAYS (Render As You See)

• Vision-Realistic Rendering Using Hartmann-Shack
  Wavefront Aberrations

Brian A. Barsky Daniel D. Garcia Stanley A. Klein
Woojin M. Yu Billy P. Chen Sarang S. Dalal
2
Motivation

• CWhatUC
• ray tracing approach similar to Kolb, but with
  corneal topography
• 3 Major Limitations
• computation time
• small light sources
• artifacts from the model

3
The Algorithm (our goal)

• color image with depth values
• fixation point and wavefront data
• generate an image with depth of field and visual
  artifacts consistent with the wavefront

4
Hartmann-Shack Device
5
Hartmann-Shack Device
6
Hartmann-Shack Device
7
(No Transcript)
8
Resampling Hartmann-Shack wavefront data

• encodes gradient/slope of wavefront
• however, sampling is very sparse!
• Solution fit a Zernike polynomial to these
  samples
• resample at higher rate

9
Point Spread Function

• 2D Retinal energy histogram
• PSF as a blur filter and subsequent convolution
  with image

10
Object-Space Point Spread Function (OSPSF)

• analog to the traditional PSF used for
  wavefronts
• image-space PSF vs. object-space OSPSF
• array of OSPSFs at each depth plane
• OSPSF as a generalization of PSF
• lens to pick fixation point
• at each depth plane, a tuned diverging lens
  converts a wavefront converging at that depth to
  a plane wave

11
Depth Planes

• A person with 20/20 vision can perceive a
  subtended angle of one minute
• ? p?D
• Assuming a typical pupil diameter of 2.4mm
• ?D 0.12 diopters
• We use ?D 0.25 diopters, but the maximum error
  is half the chosen precision
• Consider a point 1.6 meters away from the lens
• optometrists typically use 0.25 diopter
  increments for corrective lenses

12
Depth Planes

• The range of sharp human vision is 10cm to
  infinity
• At 0.25 diopter increments, we use 41 depth
  planes spaced from 0D (infinitely far) to 40D (10
  cm)
• Depth planes are placed densely closer to the
  observer, and more sparsely as it approaches
  infinity

13
Rendering Image

• PRMan
• RGBZ values

14
Stratification

• image pixels are separated according to depth
  values into depth strata
• continuous depth values are quantized into
  discrete values

15
Convolution and Accumulation

• OSPSFs are convolved with each appropriate
  depth strata
• Blurred images corresponding depth planes are
  accumulated together

16
Example Images
Cubes
Road to Pt. RAYS
17
(No Transcript)
18
Conclusion

• Vision-realistic Rendering
• Render As You See (RAYS)
• Example images