lecture 12 - mcgill cimlanger/546/12-illumination-reflectance-slides.pdf · lecture 12 illumination...
TRANSCRIPT
1
COMP 546
Lecture 12
Illumination and Reflectance
Tues. Feb. 20, 2018
Illumination and Reflectance
• Shading
• Brightness versus Lightness
• Color constancy
N(x)
L
Shading on a sunny day
𝑁
𝐿
Lambert’s (cosine) Law:
𝐼 𝑋 = 𝑁(𝑋) ∙ 𝐿
Unit Surface Normal
1
𝜕𝑍𝜕𝑋
2
+𝜕𝑍𝜕𝑌
2
+ 1
𝜕𝑍
𝜕𝑋,𝜕𝑍
𝜕𝑌,−1𝑁 ≡
𝑁
4
Shading on a sunny day
5
Cast and Attached Shadows
cast : 𝑁(𝑋) ∙ 𝐿 > 0
but light isoccluded
attached : 𝑁(𝑋) ∙ 𝐿 < 0
𝐿
Shading models
• sunny day (last lecture)
• sunny day + low relief
• cloudy day
Examples of low relief surfaces
(un)crumpled paper
Low relief surface
1
𝜕𝑍𝜕𝑋
2
+𝜕𝑍𝜕𝑌
2
+ 1
𝜕𝑍
𝜕𝑋,𝜕𝑍
𝜕𝑌,−1𝑁 ≡
9
𝜕𝑍
𝜕𝑋≈ 0
≈0 ≈0
𝜕𝑍
𝜕𝑌≈ 0
Thus,
Suppose and are both small.
Linear shading model for low relief
𝐼 𝑋, 𝑌 ≈𝜕𝑍
𝜕𝑋,𝜕𝑍
𝜕𝑌,−1 ∙ 𝐿𝑋 , 𝐿𝑌 , 𝐿𝑍
10
• shadows can still occur
• one equation per point but two unknowns
Example: curtains
𝑍 𝑋, 𝑌 = 𝑍0 + 𝑎 sin( 𝑘𝑋 𝑋 )
𝐿 𝑁
𝑋
𝑍
Example: curtains
𝑍 𝑋, 𝑌 = 𝑍0 + 𝑎 sin( 𝑘𝑋 𝑋 )
𝜕𝑍
𝜕𝑋= 𝑎 𝑘𝑋 cos 𝑘𝑋 𝑋
𝜕𝑍
𝜕𝑌= 0,
𝑋
𝑍
Example: curtains
𝐼 𝑋, 𝑌 ≈𝜕𝑍
𝜕𝑋,𝜕𝑍
𝜕𝑌, −1 ∙ 𝐿𝑋 , 𝐿𝑌 , 𝐿𝑍
𝑍 𝑋, 𝑌 = 𝑍0 + 𝑎 sin( 𝑘𝑋 𝑋 )
𝜕𝑍
𝜕𝑋= 𝑎 𝑘𝑋 cos 𝑘𝑋 𝑋
𝐼 𝑋, 𝑌 = 𝑎 𝑘𝑋 cos 𝑘𝑋 𝑋 𝐿𝑋 − 𝐿𝑍
𝜕𝑍
𝜕𝑌= 0,
Example: curtains
𝑍 𝑋, 𝑌 = 𝑍0 + 𝑎 sin( 𝑘𝑋 𝑋 )
𝐼 𝑋, 𝑌 = − 𝐿𝑍 + 𝑎 𝑘𝑋 𝐿𝑋 cos 𝑘𝑋 𝑋
𝐿 = (𝐿𝑋, 𝐿𝑌, 𝐿𝑍)
𝑋
𝑍
𝑁
Q: where do the intensity maxima and minima occur?
Shading on a cloudy day(my Ph.D. thesis)
15
(x)
Shading on a Cloudy Day
Shadowing effects cannot be ignored.
Shading on a Cloudy Day
17
Shading is determined by shadowing and surface normal.
Shading on a Sunny Day
Shading determined by surface normal only.
Shape from shading
19
Q: What is the task ? What problem is being solved?
A: Estimate surface slant, tilt, curvature.
How to account for (or estimate) the lighting ?
Illumination and Reflectance
• Shading
[Shading and shadowing models assume that intensity variations on a surface are entirely due to illumination. But surfaces have reflectance variations too. ]
• Brightness versus Lightness
• Color constancy
Which paper is lighter ?
21
Paper on the left is in shadow. It has lower physical intensity and it appears darker. But do both papers seem to be of same (white) material?
Which paper is lighter ?
22
Image is processed so that the right paper is given same image intensities as left paper. Now, right paper appears to be made of different material. Why?
𝐼 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥, 𝑦)
Abstract version
“Real” example
𝐼 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥, 𝑦)
shading & shadows material
luminance
Physical quantities
low reflectance, high illumination
low reflectance, low illumination
high reflectance, low illumination
𝐼 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥, 𝑦)
𝐼 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥, 𝑦)
“lightness”
“brightness”
Perceptual quantities
(no standard term for perceived illumination)
All four indicated “squares” have same intensity.
What is the key difference between the two configurations ?
Adelson’s corrugated plaid illusion.
“Lightness” perception
Q: What is the task ?
What problem is being solved?
A:
“Lightness” perception
Q: What is the task ?
What problem is being solved?
A: Estimate the surface reflectance, by
discounting the illumination.
𝐼 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥, 𝑦)
?
“Lightness” perception: solution sketch
Q: What is the task ?
What problem is being solved?
A: Estimate the surface reflectance, by
discounting illumination effects.
Compare points that have same illumination.
𝐼 𝑥1, 𝑦1 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥1, 𝑦1)
𝐼 𝑥2, 𝑦2 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 (𝑥2, 𝑦2)=
Illumination and Reflectance
• Shading
• Brightness versus Lightness
• Color constancy
Recall lecture 3 - color
32
Illumination
Surface Reflectance (fraction)
Absorption by photoreceptors(fraction)
There are three different spectra here.
LMS cone responses
𝐼 𝑥, 𝑦, 𝜆 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦, 𝜆 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 𝑥, 𝑦, 𝜆
Cone response
𝐼 𝑥, 𝑦, 𝜆 𝐶𝐿𝑀𝑆(𝜆) 𝑑𝜆
Surface Color Perception
34
Q: What is the task? What is the problem to be solved?
A:
illumination
Surface Reflectance
Surface Color Perception
35
Q: What is the task? What is the problem to be solved?
A: Estimate the surface reflectance, by discounting the illumination.
𝐼 𝑥, 𝑦, 𝜆 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑥, 𝑦, 𝜆 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒 𝑥, 𝑦, 𝜆
illumination
Surface Reflectance
𝐼𝑅𝐺𝐵 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝐺𝐵 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒𝑅𝐺𝐵 (𝑥, 𝑦)
For simplicity, let’s ignore the continuous wavelength 𝜆 and just consider 𝑅𝐺𝐵 (LMS).
𝜆
Cone response
𝐼 𝑥, 𝑦, 𝜆 𝐶𝐿𝑀𝑆(𝜆) 𝑑𝜆
“Color Constancy”
37
Task: estimate the surface reflectance, by discounting the illumination
𝐼𝑅𝐺𝐵 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝐺𝐵 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒𝑅𝐺𝐵 (𝑥, 𝑦)
illumination
Surface Reflectance
Why we need color constancy
• object recognition
• skin evaluation (health, emotion, …)
• food quality
• …
Example 1: spatially uniform illumination
= *
𝐼𝑅𝐺𝐵 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝐺𝐵 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒𝑅𝐺𝐵 (𝑥, 𝑦)
Given this, how to estimate this?
Solution 1: ‘max-RGB’ Adaptation
= *
𝐼𝑅𝐺𝐵 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝐺𝐵 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒𝑅𝐺𝐵 (𝑥, 𝑦)
Divide each 𝐼𝑅𝐺𝐵 channel by the max value of 𝐼𝑅𝐺𝐵 in each channel.
When does this give the correct answer?
Example 2: non-uniform illumination
= *
𝐼𝑅𝐺𝐵 𝑥, 𝑦 = 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝐺𝐵 𝑥, 𝑦 ∗ 𝑟𝑒𝑓𝑙𝑒𝑐𝑡𝑎𝑛𝑐𝑒𝑅𝐺𝐵 𝑥, 𝑦
Solution: See Exercises.
sun +blue sky
onlyblue sky
Illumination and Reflectance
• Shape from Shading
• Brightness versus Lightness
• Color constancy
Different solutions require different underlying assumptions. This is separate issue from how we can code up a solution using neural circuits.