Chapter Overview: Point Operations!: Relating Gray Values To Brightness!

Chapter Overview: Point Operations!
  Relating gray values to brightness!

  Gray value quantization!
  Weberʼs Law!
  Gamma characteristic!
  Adjusting brightness and contrast!
  Gray-level histograms and histogram equalization!
  Point operations for combining images!
  Averaging!
  Subtraction!
  The need for image registration!
Bernd Girod: EE368 Digital Image Processing! Point Operations no. 1!

Quantization: how many bits per pixel?!
8 bits
5 bits
4 bits

„Contouring“

3 bits
2 bits
1 bit


How many gray levels are required?!
  Contouring is most visible for a ramp!
32 levels!
64 levels!
128 levels!
256 levels!
  Digital images typically are quantized to 256 gray levels. !

Brightness discrimination experiment!
  Can you see the circle?!
I + ΔI
Note: I is luminance,
125!
126!
128!
129!
130!
127!
I measured in cd m 2
  Visibility threshold!
„Weber fraction“!
ΔI I ≈ KWeber ≈ 1…2% „Weberʻs Law“!

Contrast ratio without contouring!
  Luminance ratio between two successive quantization
levels at visibility threshold!
I max N −1
= (1 + KWeber )
I min
I max
  For! KWeber = 0.010.02 N = 256 = 13156
I min
  Typical display contrast ratio!
  Modern flat panel display in dark room 1000:1!
  Cathode ray tube 100:1!
  Print on paper 10:1!
  Suggests uniform perception in the log(I) domain
(„Fechnerʻs Law“)!

Gamma characteristic!
  Cathode ray tubes (CRTs) are nonlinear!

Luminance!
I!
I ~ Uγ
γ = 2.0 . . . 2.3!
Voltage U!
  Cameras contain γ-predistortion circuit!

U ~ I1 γ

log vs. γ-predistortion!
U

U ~ log(I )
U ~ I1 γ
Imax
= 100
Imin
  Similar enough for most practical applications!

Photographic film!
Hurter & Driffield curve (H&D curve)!
for photographic negative!
slope -γ! Luminance!

2.0
shoulder!
I = I 0 ⋅10 − d
= I 0 ⋅10 −( − γ log E +d0 )
density d

„linear“ region!
1.0
= I 0 ⋅10 − d0 ⋅ E γ
toe! log E

0

d0
E is exposure

  γ measures film contrast!
  General purpose films: γ = -0.7 . . . -1.0!
  High-contrast films: γ = -1.5 . . . -10!
  Lower speed films tend to have higher absolute γ!

Brightness adjustment by intensity scaling!
Original image! Scaled image!
f [ x, y ] a ⋅ f [ x, y ]
Scaling in the γ-domain is equivalent to scaling in the linear luminance domain 

γ γ
I ~ ( a ⋅ f [ x, y ]) = aγ ⋅ ( f [ x, y ])
. . . same effect as adjusting camera exposure time.!

Contrast adjustment by changing γ!
Original image! γ increased by 50%!
f [ x, y ] γ
a ⋅ ( f [ x, y ]) with γ = 1.5
. . . same effect as using a different photographic film . . .!
Contrast adjustment by changing γ!
Scaled ramp 2γ0!
Original ramp γ0!
Scaling chosen to!

approximately preserve!
brightness of mid-gray!
Scaled ramp 0.5γ0!
Gray level histograms!
  Distribution of gray levels can be judged by measuring a 

histogram:!
  For B-bit image, initialize 2B counters with 0 !
  Loop over all pixels x,y!
  When encountering gray level f [x,y]=i, increment counter #ι

  Normalized histogram may be thought of as an 
empirical probability distribution.!
  You can also use fewer, larger bins to trade off amplitude 
resolution against sample size.!
#pixels! Example histogram!
Cameraman!
image!
gray level!
#pixels!
Example histogram!
Pout!
image!
gray level!
Histogram equalization!
  Idea: find a non-linear transformation!

g=T(f)
!to be applied to each pixel of the input image f [x,y], such 
that a uniform distribution of gray levels in the entire range 
results for the output image g[x,y].

  Analyse ideal, continuous case first, assuming!
  0
≤ f ≤ 1 0 ≤ g ≤1
  T(f) is strictly monotonically increasing, hence, there
exists!
f = T −1 ( g ) 0 ≤ g ≤1
  Goal: pdf pg(g) = 1 over the range !0 ≤ g ≤ 1
Histogram equalization for continuous case!
  From basic probability theory!

⎡ df ⎤
pg ( g ) = ⎢ p f ( f ) ⎥
⎣ dg ⎦ f =T −1 ( g )
  Consider the transformation function!
f
g = T ( f ) = ∫ p f (α ) dα 0 ≤ f ≤1
0
  Then . . . ! dg
= pf ( f )
df
⎡ df ⎤ ⎡ 1 ⎤
pg ( g ) = ⎢ p f ( f ) ⎥ = ⎢ p f ( f ) ⎥ =1 0 ≤ g ≤1
⎣ dg ⎦ f =T −1 ( g ) ⎢⎣ p f ( f ) ⎥⎦ −1
f =T ( g )
Histogram equalization for discrete case!
  Now, f only assumes discrete amplitude values f0 , f1 ,, fL−1!

!with empirical probabilities!
L−1
n0 n1 nL−1
P0 = P1 =  PL−1 = where n = ∑ nl
n n n l =0
f
  Discrete approximation of! g = T ( f ) = ∫ p f (α ) dα
0
k
gk = T ( fk ) = ∑ Pi for k = 0,1..., L − 1
i=0
  The resulting values gk are in the range [0,1] and need to

be scaled and rounded appropriately. !
Histogram equalization example!
Original image Pout! Pout!

after histogram equalization!
Original image Pout! . . . after histogram equalization!
#pixels!
#pixels!
gray level! gray level!
Original image Cameraman!

Cameraman! after histogram equalization!
Original image Cameraman! . . . after histogram equalization!
#pixels!
#pixels!
Original image Moon! Moon!

after histogram equalization!
Original image Moon! . . . after histogram equalization!
#pixels!
#pixels!
Contrast-limited histogram equalization!
Equalization with!
original histogram!
Output gray level!

#pixels!
Equalization with!
clipped histogram!
Histogram Clipping!
Input graylevel!
gray level! Input gray level!
Adaptive histogram equalization!
  Apply histogram equalization based on a histogram obtained
from a portion of the image!
Sliding window approach:  Tiling approach: 

different histogram (and mapping)  subdivide into overlapping regions, 
for every pixel! mitigate blocking effect by smooth blending 
between neighboring tiles !
  Must limit contrast expansion in flat regions of the image, 

e.g. by clipping individual histogram values to a maximum 
(CLAHE - “Contrast-limited adaptive histogram equalization”)!
[Pizer, Amburn et al. 1987]!
Original! Global histogram! Tiling  Tiling 

8x8 histograms! 32x32 histograms!
Original image Dental Xray! Adaptive histogram equalization, 16x16 tiles!
Dental Xray after  

equalization of 
global histogram!
Original image! Global  Adaptive histogram!

histogram equalization 
equalization! 8x8 tiles!
Point operations for combining images!
  Image averaging for noise reduction!

  Combination of different exposures for 
high-dynamic range imaging!
  Image subtraction for change detection!
  Accurate alignment is always a requirement!
Image averaging for noise reduction!
1 image 2 images 4 images!
http://www.cambridgeincolour.com/tutorials/image-averaging-noise.htm!
High-dynamic range imaging!
16 exposures, one f-stop (2X) apart!
[Debevec, Malik, 1997] ! Combined image!

Image subtraction!
  Find differences/changes between 2 mostly identical images!

  Example: digital subtraction angiography!
- +
Contrast!
enhancement!
http://www.isi.uu.nl/Research/Gallery/DSA/!
Video background subtraction!
New Frame! Background Frame! New Frame! Background Frame!
+! -! +! -!
Abs(.) > θ ?
Abs(.) > θ ?

Update: Background[t] := α Background[t-1] + (1- α) New[t]

Image subtraction example from IC manufacturing: 
die-to-die comparison of photomasks!
Where is the defect?!
Image A (no defect)! Image B (w/ defect)!
Absolute difference between two images!
w/o alignment! w/ alignment and thresholding!

Chapter Overview: Point Operations!: Relating Gray Values To Brightness!

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Chapter Overview: Point Operations!: Relating Gray Values To Brightness!

Caricato da

Copyright:

Formati disponibili

Chapter Overview: Point Operations!

 Relating gray values to brightness!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 1!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 2!

 Contouring is most visible for a ramp!

 Digital images typically are quantized to 256 gray levels. !

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 3!

 Can you see the circle?!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 4!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 5!

 Cathode ray tubes (CRTs) are nonlinear!

 Cameras contain γ-predistortion circuit!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 6!

 Similar enough for most practical applications!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 7!

slope -γ! Luminance!

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 8!

Original image! Scaled image!

Scaling in the γ-domain is equivalent to scaling in the linear luminance domain

Bernd Girod: EE368 Digital Image Processing! Point Operations no. 9!

Original image! γ increased by 50%!

. . . same effect as using a different photographic film . . .!

Scaled ramp 2γ0!

Original ramp γ0!

Scaling chosen to!

 Distribution of gray levels can be judged by measuring a

 Idea: find a non-linear transformation!

 Goal: pdf pg(g) = 1 over the range !0 ≤ g ≤ 1

 From basic probability theory!

 Now, f only assumes discrete amplitude values f0 , f1 ,, fL−1!

 The resulting values gk are in the range [0,1] and need to

Original image Pout! Pout!

Original image Cameraman!

Original image Moon! Moon!

gray level! gray level!

Output gray level!

Sliding window approach: Tiling approach:

 Must limit contrast expansion in flat regions of the image,

Original! Global histogram! Tiling Tiling

Original image Dental Xray! Adaptive histogram equalization, 16x16 tiles!

Dental Xray after

Original image! Global Adaptive histogram!

 Image averaging for noise reduction!

16 exposures, one f-stop (2X) apart!

[Debevec, Malik, 1997] ! Combined image!

 Find differences/changes between 2 mostly identical images!

Update: Background[t] := α Background[t-1] + (1- α) New[t]

Image A (no defect)! Image B (w/ defect)!

w/o alignment! w/ alignment and thresholding!

Potrebbero piacerti anche

  Relating gray values to brightness!

  Contouring is most visible for a ramp!

  Digital images typically are quantized to 256 gray levels. !

  Can you see the circle?!

  Cathode ray tubes (CRTs) are nonlinear!

  Cameras contain γ-predistortion circuit!

  Similar enough for most practical applications!

Scaling in the γ-domain is equivalent to scaling in the linear luminance domain 

  Distribution of gray levels can be judged by measuring a 

  Idea: find a non-linear transformation!

  Goal: pdf pg(g) = 1 over the range !0 ≤ g ≤ 1

  From basic probability theory!

  Now, f only assumes discrete amplitude values f0 , f1 ,, fL−1!

  The resulting values gk are in the range [0,1] and need to

Sliding window approach:  Tiling approach: 

  Must limit contrast expansion in flat regions of the image, 

Original! Global histogram! Tiling  Tiling 

Dental Xray after  

Original image! Global  Adaptive histogram!

  Image averaging for noise reduction!

  Find differences/changes between 2 mostly identical images!