Dynamic Range Reduction for Tone Mapping of HDR Images

1
Dynamic Range Reduction for Tone Mapping of HDR
Images
Sumanta Pattanaik
2
HDR Tone Mapping Problem
the absolute range of environmental radiances
is vast (gt 8 log units)
100
3
HDR Tone Mapping Problem
the dynamic range of radiances in a scene can
also be large (gt4 log units)
100
4
HDR Tone Mapping Problem
the absolute and dynamic range of display
device is small (1.5 - 2.0 log units)
100
Range of Typical Displays from 1 to 100 cd/m2
0 255
5
HDR Tone Mapping Problem
How do we compress an HDR image so as to
realistically display on a LDR display device?
Image Source Garettt Johnson, RIT
6
HDR Range Compression
7
HDR Range Compression
8
HDR Range Compression
9
Visual System and Range Compression
Visual Adaptation Local Adaptation
Multi-scale Adaptation
10
Visual Adaptation
Psychophysical Studies
LaDL
La
Webers Law DLkLa. Over a wide range of
ambient light, La, DL is directly proportional to
La.
11
Visual Sensitivity
1 Visual
Sensitivity --------
DL Sensitivity is inversely
proportional to the adaptation luminance La.
12
Visual Sensitivity
1 Visual
Sensitivity --------
DL Sensitivity is inversely
proportional to the adaptation luminance La.
13
Sensitivity based Gain Control

• Gain proportional to visual sensitivity.

14
Sensitivity based Gain Control
Cone dominated
Gain
rod
cone
log Gain
1000 cd/m2
6
-2
-6
0
2
4
-4
log La
15
Sensitivity based Gain Control
Rod dominated
0.04 cd/m2
16
Sensitivity based Gain Control
Ward 1994
Mathematical manipulation will show that the
above equation is the same as
17
Sensitivity based Gain Control
Ferwerda 1996
18
Visual Adaptation
Physiological Studies

• Retina
• Converts light in to signals
• Transmits signal to brain

Light
19
Retinal Response

• Cells of the visual system have limited response
  range.

Rmax
n
0.5?Rmax
?
-5
-4
-3
20
Retinal Response to Various La
R / Rmax
In Dark
?0
luminance L
21
Retinal Response to Various La
R / Rmax
BackgroundLevel L1
?1
luminance L
22
Retinal Response to Various La
R / Rmax
BackgroundLevel L2
?2
luminance L
23
Retinal Response to Various La
BackgroundLevel L3
R / Rmax
?3
luminance L
24
Retinal Response to Various La
BackgroundLevel L4
R / Rmax
?4
luminance L
25
Retinal Response to Various La
Adapted from The retina An approachable part
of the brain by John E. Dowling, Belknap press.
(page 87)
R / Rmax
luminance L
26
Retinal Response Model
27
Retinal Response Model
Pattanaik 2000, Reinhard 2002, Reinhard 2005
28
Visual Adaptation Model
Photoreceptor Response Model
Threshold Sensitivity Model
29
How to Computer La?

• Image average (Global)
• Arithmetic average
• Geometric average
• Image average (Local)
• Gaussian filtering
• Multi-scale averages
• Bilateral Filtering

30
Image Average (Global)

• Arithmetic average

31
Image Average (Global)

• Geometric average

32
Image Average (Global)

• Geometric average

33
Image Average (Global)

• Geometric average

or
34
Global Average Is not Useful for HDR images
35
Local Average

• our eyes adapt as we look from place to place in
  a scene
• local adaptation enhances our ability to see in
  high dynamic range scenes
• our visual impression of the scene is constructed
  from what we are able to see as we look around

36
Local Average

• Filtering.

Box filter Gaussian filter
37
Original
Box Filter
Gaussian Filter
38
Result using a Local Filter
Box Filter
39
Local Average

• Bilateral Filtering.

Box filter
Gaussian filter
40
Local Average

• Bilateral Filtering.

Tapered Box filter
Gaussian filter
Durand 2002 Bilateral Filter
Pattanaik 2002 Susan Filter
41
Result using Local average from Bilateral Filter
Bi-Box Filter
42
Original
Bi-Box Filter
Bi-Gaussian Filter
43
Original
Box Filter
Gaussian Filter
44
Effect in Tone mapping
Box Filter
Bi-Box Filter
45
Local Average

• Adaptive Kernel size
• Similar in spirit to Bilateral filtering.
• Depending on how the center pixel value differs
  from the surround larger or smaller size is
  chosen.

Used by Reinhard 2002 Ashikhmin 2002
46
Local Average

• Multiscale Approach

_
_
_
47
Center-Surround mechanisms
Center
Surround
48
Local Average

• Multiscale Approach

Pattanaik 1998 Surround provides
the La.
Li 2005 Difference of center surround
provides the La.
49
Summary

• Visual Adaptation Based Range Compression
• Local Adaptation for Realistic appearance