25-07-2022

Photometric model

Digital Image Processing • Suppose light intensity I is transported ideally to image plane 𝑥, 𝑦 .
• Scene reflectance R is also transferred appropriately to image plane.
and Analysis • Thus image intensity mapped is ideal one and is given by
(Digital image fundamentals) 𝑓 𝑥, 𝑦 = R 𝑥, 𝑦 I 𝑥, 𝑦

• However, light energy transport to real image plane undergoes some

Bhabatosh Chanda transformation. This may be modeled by photometric transformation.
𝑔 𝑥, 𝑦 = T[𝑓 𝑥, 𝑦 ]
bchanda57@gmail.com
that satisfies the condition: 𝑓 𝑥, 𝑦 ≥ 0 and 𝑔 𝑥, 𝑦 ≥ 0

Outline Photometric transformation model

• Image formation
• Geometric transformation • Three basic principles to follow:
• Photometric transformation
• Non-negativity
• Digital image fundamentals 𝑓 𝑥, 𝑦 ≥ 0 and 𝑔 𝑥, 𝑦 ≥ 0
• Digitization
• Colour model • Neighbourhood process
• Image representation • Superposition
• Relevant digital geometry

7/25/2022 2

1
 25-07-2022

Photometric model Image formation equation

• Assumptions: • In discrete domain, for M  N digital image
N 1 M 1
• Photometric transformation T is defined as convolution with a g (r , c )  
function h(x,y) that is linear and shift invariant plus some noise. j 0

i 0
f (i, j )h(r  i, c  j )   (r , c)

• Noise is signal-independent and additive. This represents 2-dimensional convolution.

• Original and observed image intensity is non-negative. • Using matrix-vector notation
g  Hf  
This represents 𝑀𝑁 linear equations
• Degradation matrix is of size 𝑀𝑁 × 𝑀𝑁.

Digital image Digital graylevel image g (r , c)

Discrete domain D  Z , where

2

• In the analog domain

D  {( r , c ) | r  0,1,2,........., M  1; c  0,1,2,........, N  1}

𝑔(𝑥, 𝑦) = 𝑓 𝛼, 𝛽 ℎ 𝑥 − 𝛼, 𝑦 − 𝛽 𝑑𝛼 𝑑𝛽 + 𝜂(𝑥, 𝑦)  g ( 0, 0 ) g (0,1) ..... ..... g (0, N  1) 

 g (1,0) g (1,1) ...... ...... g (1, N  1) 

g (r , c)   ...... ...... ...... ...... ...... 
• Corresponding digital image of size 𝑀 × 𝑁 is  
 ...... ...... ...... ...... ...... 
 g ( M  1,0) g ( M  1,1) ...... ...... g ( M  1, N  1) 
𝑔(𝑥, 𝑦 = 𝑓 𝑚, 𝑛 ℎ 𝑥 − 𝑚, 𝑦 − 𝑛 + 𝜂(𝑥, 𝑦)
where g ( r , c )  0,1, 2,......... ..., L  1

7/25/2022 8

2
 25-07-2022

Colour models Bayer filter

• Bayer filter is a color filter array used
• Colour is a complex perceptual phenomenon. with sensor of digital camera.
• Sensation of colour arises due to response of three neurochemical • Arranged in square grid with 50% G,
25% R and 25% B.
sensors or receptors in the retina to the visible light.
• Each pixel gets one particular color
R   C ( ) hR ( ) d that the filter allows to reach the
sensor.
• Other color components are
G   C ( ) hG ( ) d interpolated (averaged) from the
neighborhood.
B   C ( ) hB ( ) d

7/25/2022 9 7/25/2022 Computer Vision -- Intro 11

Retina and camera sensor RGB colour model

• A unit cube in Cartesian

coordinate system.
• Additive primaries, i.e.,
R G  B W
• Used in devices like camera,
monitor.
• (0,0,0)  Black
• (1,1,1)  White

7/25/2022 10 7/25/2022 12

3
 25-07-2022

CMY colour model Hue: mixing RGB

• Complementary to RGB colour model, so uses same system for
representation.
• Subtractive primaries, i.e.,
Cyan = 1 – red
Magenta = 1 – green
Yellow = 1 – blue
• Used in printing devices.
• (0,0,0)  white and (1,1,1)  black

7/25/2022 13 7/25/2022 15

HLS colour model Saturation: Purity

• Model is represented by a double-

hex cone.
• User oriented model (for interaction)
based on intuitive colour notion.
0o  H  360o (circular)

0  S  100

0 V 1

7/25/2022 14 7/25/2022 16

4
 25-07-2022

Saturation: Purity Example of colour image

• Hue shows pure colours are formed by combining two of 109 106 106 110 111 106 60 98 142 114 94 98 102 120 103 109
115 126 126 137 137 136 106 146 157 146 124 128 98 96 89 82
the additive primaries 137 118 134 145 151 147 142 157 154 140 120 105 85 77 67 63
189 150 123 131 145 143 143 134 123 110 90 65 65 67 83 102
193 191 157185 182 161 126 170 188 181 167 158 153 151 155 155 158 160 163

• Minimum of R, G and B combines with others to form 193 189 185185 178 147 125 171 183 192 181 167 157 155 160 152 156 160 160
168 103 74178 179 145 139 163 171 190 195 176 164 161 162 162 150 154 155
90 106 112175 177 148 150 163 169 186 197 191 169 167 167 160 158 150 152
whiteness in colour, e.g., pink = red + white 125 133 136164 169 142 142 114 94 98 102 120 103 109 185 182 161 126 170 188 181 167
143 154 165154 171 172 185 178 147 125 171 183 192 181 167 157 155 160 152 156 160 160
170 185 187182 190 189 178 179 145 139 163 171 190 195 176 164 161 162 162 150 154 155
• Saturation suggests the purity of colour 180 179 174174 166 168 175 177 148 150 163 169 186 197 191 169 167 167 160 158 150 152
142 116 128169 165 162 164 169 142 154 171 172 182 190 189 174 166 168 169 165 162 150
141 148 154150 163 165 163 165 139 153 175 163 179 189 187 178 166 167 170 169 170 168
160 163 164139 153 175 162 161 136 144 164 163 171 189 176 179 169 169 169 164 163 169
168 171 181163 179 189 109 106 106 110 111 106 60 98 142 114 94 98 102 120 103 109
187 178 166 115 126 126 137 137 136 106 146 157 146 124 128 98 96 89 82
167 170 169 157 146 124 128 98 96 89 82 193 189 185 168 103 74 90 106
170 168 162 112 125 133 136 143 154 165 170 154 140 120 105 85 77 67 63
161 136 144 185 187 180 179 174 142 116 128 141 148 154 160 163 164 168 171
181 185 178 173 179 171 162 153 153 160 162 164 164 163 165 167
183 184 175 149 177 179 165 164 155 153 159 164 160 159 163 164
154 140 120 105 85 77 67 63 123 110 90 65 65 67 83 102
95 98 84 94 107 119 138 155 112 125 133 136 143 154 165 170

7/25/2022 17 7/25/2022 19

RGB to YCbCr Digital colour image g C ( r , c)

Discrete domain D  Z , where

2

• Present days all digital videos and static images are compression
standards are based on D  {( r , c ) | r  0,1,2,........., M  1; c  0,1,2,........, N  1}
𝑌 0.299 0.587 0.114 𝑅′ 0  g C ( 0, 0) g C (0,1) ..... ..... g C (0, N  1) 
𝐶 = −0.168736 −0.331264 0.5 𝐺′ + 128  g (1,0) g C (1,1) ...... ...... g C (1, N  1) 
𝐶  C
0.5 −0.418688 −0.81312 𝐵′ 128 [ g C ( r , c )]MxN  ...... ...... ...... ...... ...... 
 
...... ...... ...... ...... ......
• R’, G’, B’ are gamma-corrected R, G, B.  
 g C ( M  1,0) g C ( M  1,1) ...... ...... g C ( M  1, N  1)
• Y is intensity component. Others represent colors.
where g C (r , c )  [ g R (r , c) g G (r , c) g B (r , c) ]
T

and g R ( r , c ), g G ( r , c ) and g B ( r , c )  0,1,2,......... ..., L  1

7/25/2022 18 7/25/2022 20

5
 25-07-2022

Types of colour Pseudo Colour

Three different approaches are adopted to generate colour images for
better visualization.
1. True colour or natural colour
2. Pseudo colour
3. False colour composite

7/25/2022 Workshop on Image Processing and Synthesis 21 7/25/2022 Workshop on Image Processing and Synthesis 23/129

Pseudo Colour Pseudo Colour

• Convert graylevel image to colour image.
• Different graylevels are mapped to different order.
• Order of intensity is preserved in terms of hue

7/25/2022 Workshop on Image Processing and Synthesis 22

6
 25-07-2022

Pseudo Colour False colour composite

Band-1 Band-2

7/25/2022 Workshop on Image Processing and Synthesis 27/129

False colour composite False colour composite

• Assigns different colours to different bands of multichannel image
Band-3 False colour composite
data.
• Most used in combining satellite imagery.
• Channel bandwidth and colour assigned may not be related.
• Purpose is to increase distinctness among different image features.

7/25/2022 Workshop on Image Processing and Synthesis 26 7/25/2022 Workshop on Image Processing and Synthesis 28/129

7
 25-07-2022

Gamma correction
• Response of photosensor to input
is non-linear (red curve).
𝐵= 𝑉

• That we may compensate by

applying inverse operation (blue
curve).
•𝐼=𝐵 /
= 𝑉 / = V  linear

7/25/2022 29 7/25/2022 Workshop on Image Processing and Synthesis 31/129

Gamma correction

7/25/2022 30