You are on page 1of 50

Image Enhancement in the

Spatial Domain II
Review: gray level transformations

For this image,


what transform
should we apply?
Histogram Processing
 Histogram equalization
 Histogram matching
 Local enhancement
 Use of histogram statistics
Histogram
 Statistics of the pixel gray-levels of an image
 h(rk)=nk : histogram
no. of
occurrence

gray level
Obtain contrast information
from histogram
Histogram: implementation issue
 Divide the range of gray levels (ex. 0~255)
 Number of bins
 Position of bins specified in a vector

10 bins
5 bins
Histogram: normalization

no. of no. of occurrence


=prob.
occurrence total pixels

gray level gray level


Histogram equalization 均化
 Goal: produce an uniform histogram
 Find a transformation: s=T(r)

uniform

0 1 0 1
Histogram equalization(cont.)
Cumulative function of
1 histogram

0.5 Ideal cumulative func.

Prob. 0.5

0.5

0 1 0 0.5 1
Ideal uniform histogram
Histogram equalization:
discrete solution
 Probability (normalized histogram) of gray
level rk
 p(rk)=nk/n, k=0,1,2,…, L-1
 Discrete version of CDF (cumulative
distribution function)
k k nj
sk  T (rk )   pr (rj )   <1
j 0 j 0 n
Example: histogram equalization
Histogram Processing
 Histogram equalization
 Histogram matching
 Local enhancement
 Use of histogram statistics
Histogram matching: original
image
Histogram equalization result
Histogram
Equalization
function

Equalized
histogram
After equalization
Histogram matching
 The desired shape of histogram is specified
(not necessarily uniform)
 Derivation of histogram matching function:

z=G-1(T(r))
z s r
desired uniform input
z=G-1(s) s=T(r)
histogram
equalization

histogram
equalization
s=G(z)
Original -> uniform desire -> uniform
(equalization) (equalization)

Inverse
function
of G
Histogram matching result
Desired histogram

After histogram
matching

G-1
Histogram Processing
 Histogram equalization
 Histogram matching
 Local enhancement
 Use of histogram statistics
Local enhancement
 Global: entire image
 Local: neighborhood of each pixel

Original image Global equalization 7x7 neighborhood


local equalization
Histogram statistics
 Take pixel value r as a random variable
Normalized histogram of an image
=> probability distribution function of r

 Some measure about pdf


L 1
m   ri p ( ri ) mean
i 0
L 1
 2 (r )   (ri  m) 2 p (ri ) variance
i 0

mean => average gray level


variance => average contrast
Histogram statistics (cont.)
 Local measurement: Sxy neighborhood

mS xy  r
( s ,t )S xy
s ,t p(rs ,t ) mean in Sxy

 (r ) 
2
S xy  (r
( s ,t )S xy
s ,t  mS xy ) p(rs ,t )
2
variance in Sxy
Application of histogram statistics:
example
Globally, this image is light

However, some dark


background object is
unseen

鎢絲及其支柱的放大影像
Histogram statistics: example
Calculate local statistics in 3x3 region
local mean local standard deviation
Histogram statistics: example
 Local enhancement 像素平均值小 但是標準差大
E  f ( x, y )k1 if mS xy  k0 M G and DG   S xy  k1 DG
g ( x, y )  
 f ( x, y ) otherwise

enhanced
pixels
Histogram statistics: example

original enhanced
Outline: spatial domain
 Histogram processing
 Spatial filtering
 Smoothing filter
 Ordered statistics filter
 Sharpening filter
Basics of spatial filtering

 g(x,y)=T[ f(x,y) ]
 T operates on a
neighborhood
mask coefficients

X (product) output

underlying neighborhood
Basics of spatial filtering
 Mathematical formula
a b
g ( x, y )    w(s, t ) f ( x  s, y  t )
s   at   b
convolution

 Border issues
 Operate inside only
 Discard the mask at border
 Padding at image boundaries
 zero padding
 extension
Smoothing spatial filtering
 Used for blurring and for noise reduction
 Smoothing linear filters
 Average filters, low-pass filters

average filter weighted filter 這兩個效果差不多


因為3x3太小
original 3x3

noise
reduced

5x5 9x9

small area
blended into
background

15x15 35x35
border
effects
Smoothing: example

15x15 smoothing thresholding


Order-statistics filters
 Pixel neighborhood
f1 f2 f3
f4 f5 f6
f7 f8 f9
sort
f4 f3 f7 f6 f8 f2 f1 f9 f5
increasing order

min median max


filter filter filter
remove isolated
high(low) pixels
Median filter: example

original 3x3 average 3x3 median

salt-and-pepper noise
Outline: spatial domain
 Histogram processing
 Spatial filtering 空間濾波
 Smoothing filter
 Ordered statistics filter
 Sharpening filter
Sharpening
filtering:
example
original Edge detection

scaled sharpened
Sharpening spatial filters
 Goal
 highlight fine detail in an image
 Spatial differentiation 微分
 Digital implementation => difference 差分
 de-emphasize regions with slowly varying gray
levels
smooth

 1  ( 1) '

 2  ( 1)2 ''
Sharpening: foundation
 First and second-order derivatives in digital
form => difference
f
 f ( x  1)  f ( x) 一次微分
x
2 f
 [ f ( x  1)  f ( x)]  [ f ( x)  f ( x  1)] 二次微分
x 2

 f ( x  1)  f ( x  1)  2 f ( x)
2nd derivatives for image
 2-D 2nd derivatives => Laplacian
 2
f  2
f
 f  2  2
2

x y
=>discrete formulation
 2 f  [ f ( x  1, y )  f ( x  1, y )  2 f ( x, y )]
 [ f ( x, y  1)  f ( x, y  1)  2 f ( x, y )]
 [ f ( x  1, y )  f ( x  1, y )  f ( x, y  1)  f ( x, y  1)]  4 f ( x, y )
Definition of 2nd derivatives in
filter mask
450 rotation
900 rotation invariant
invariant (include
Diagonals)

- - - -

- 4 - - 8 -

- - - -
Laplacian filtering: example

original Laplacian filtered image


Sharpened result = original +
Laplacian filtered image
 f ( x, y)   2 f ( x, y) if center coeff. of the mask is negative
g ( x, y)  
 f ( x, y )   2
f ( x, y) if center coeff. of the mask is positive

original image filtered image

=> resulting mask


Laplacian
filtering:
example

original Laplacian

scaled sharpened
original

right
left
mask
mask
Use of 1st derivative: Gradient
 Gradient  f 
Gx   x 
f      f 
G y   
 y 

 Magnitude of gradient
2 1/ 2
 f   f  
2

f  mag (f )       
 x   y  

Simplified form => f  G x  G y


Definition of 1st derivative in
mask filters
 Sobel operators
Gradient: example
*enhance defects and eliminate slowly changing background

original(contact lens) Sobel gradient


defects