High Dynamic Range Imaging

1
High Dynamic Range Imaging

• Greg Ward

2
Abstract

• This paper will give us how to calculate full
  spectral radiance at a point and convert it to a
  reasonably correct display color.

3
Introduction

• The dynamic range of sRGB in display is only
  about 901 while human observers can perceive 4-5
  orders of magnitude in luminance.
• sRGB gamut only covers about half of perceivable
  colors.

4
Introduction

• How to solve this problem? Where to begin?
• 1. physically-based rendering.
• 2. multiple-exposure image capture.

5
Introduction

• Physically-based rendering.
• ? We will focus on in this paper.
• Multiple-exposure image capture.
• ? Read references for details.

6
Introduction

• What is rendering?
• Why we need rendering? Can we just measure all
  the optical spectrum we received?

7
Introduction

• The spectral rendering Eq.
• What is the cos? term?

8
Introduction

• Participating Media
• What is the 4p term?

9
Introduction

• Solving the Rendering Eq.
• ? frustrated.
• Three approaches
• 1. local illumination approximation.
• 2. ray tracing.
• 3. radiosity.

10
Introduction

• Local illumination approximation.
• ? convert the integral eq. over a simple sum
  over light sources.
• Ray tracing.
• ? traces additional rays to determine specular
  reflection and transmission.
• Radiosity.
• ? surface become a large linear system.

11
Introduction

• We assume that color accuracy is an important
  goal.
• Use ray tracing or radiosity.

12
Tone Mapping

• Tone Mapping
• Map each tristimulus value into our target
  displays color space.
• Two topics
• 1. luminance
• 2. gamut mapping

13
Tone Mapping

• Convert spectrum to absolute XYZ color.
• X 683?F(?)x(?)d?
• Y 683?F(?)y(?)d?
• Z 683?F(?)z(?)d?

14
Tone Mapping

• If we directly display the color A and B to the
  screen, they would appear incorrect.

15
Tone Mapping

• We need make a transform.

16
Tone Mapping

• Examples

17
Tone Mapping

• We must consider the color difference caused by
  different illuminant.

18
Tone Mapping

• The 1st row transform to its 2nd row RGB
  equivalents by matrix below.

19
Tone Mapping

• In 2nd row, B is 22?E from D 65 A is 80 ?E
  from D65
• After we made the tone mapping (In 3rd row)
• , B is only 1 ?E from D65 while A is only 5
  ?E from D65.

20
Tone Mapping

• How to perform the mapping?
• Rw , Gw , Bw The illuminant B or A after
• transform.
• Rw , Gw , Bw The reference illuminant
• D65 after
  transform.

21
Tone Mapping

• An idea of domain switching.

22
Tone Mapping

• Why dont we just use XYZ domain to make the
  approach but use M domain?
• How we determine the matrix M?

23
Tone Mapping

• We can attach C709 matrix with the transform
  matrixes and obtain an matrix to direct mapping
  2nd row to 3rd row.

24
Tone Mapping

• What if we encounter different light sources?
• We can compare them to the light sources we have.
• There maybe error associated with sources having
  diff. colors, but these will negligible in most
  scenes.

25
Tone Mapping

• Now we finish the adjust for luminance, then save
  the data for diff. display applications.
• 3 types of global tone Mapping.

26
Gamut Mapping

27
Conclusion

• Use a global illumination method for all the
  phenomena being simulated.
• Choose a good chromatic adaptation model.
• Substitute full spectral rendering with a
  relative color approximation.
• Record images in a HDR format to preserve display
  option.
• Base tone-mapping and gamut-mapping on specific
  goals.

28
Brainstorm

• Mr. Greg Ward derive the M reference from one
  object and 2 diff. lights, is it possible to
  derive a reference for more objects? Will M still
  the same?
• If we can find out how the M matrix is derived,
  maybe we can find a better matrix to perform this
  job. (or even better auto adjust matrix value.)

29
Thanks for your comments.