Lecture5 PDF

EECS490: Digital Image Processing
Lecture #5
• Point transformations (cont.)

• Histogram transformations
– Equalization
– Specification
– Local vs. global operations
• Intro to neighborhoods and spatial filtering
Brightness & Contrast
© 2002 R. C. Gonzalez & R. E. Woods

Point Ops via Functional Mappings
Image: I , point
J J = [I ]
operator
Input Output
If I (r, c )= g
function,
Pixel: I(r,c) f J(r,c)
and f (g )= k
The transformation of image I into image J is then J (r, c )= k .
accomplished by replacing each input intensity, g, with
a specific output intensity, k, at every location (r,c)
where I(r,c) = g.
The rule that associates k with g is usually
specified with a function, f, so that f (g) = k.
1999-2007 by Richard Alan Peters II

One-band Image J (r, c )= f (I (r, c )),

for all pixels locations (r, c ).
Three-band Image J (r, c, b)= f (I (r, c, b)), or

J (r, c, b)= f b (I (r, c, b)),
for b = 1,2,3 and all (r, c ).

One-band Image J (r, c )= f (I (r, c )),

Either all 3 bands for all pixels locations (r, c ).
are mapped through
the same function,
f, or …
Three-band Image J (r, c, b)= f (I (r, c, b)), or
J (r, c, b)= f b (I (r, c, b)),
for b = 1,2,3 and all (r, c )…. each band is
mapped through
a separate func-
tion, fb.

A look-up table (LUT) … then the LUT

implements a that implements f
functional mapping. is a 256x1 array
whose (g +1)th
value is k = f (g).
If k = f (g ),
for g = 0,…, 255,
To remap a 1-band
and if k takes on
image, I, to J :
values in {0,…, 255}, … J = LUT(I + 1)

If I is 3-band, then
a) each band is mapped separately using the same
LUT for each band or
b) each band is mapped using different LUTs – one
for each band.
a) J = LUT(I + 1), or
b) J (:,:, b)= LUTb (I (:,:, b)+ 1) for b = 1,2,3.


255
E.g.: index value

... ...
output value
101 64
102 68
127
103 69
104 70
105 70
106 71
0
0 127 255 ... ...

input value input output

input output
...0 ... 0
cell index
contents
64
...32
a pixel with ...
is mapped to
this value this value
128 128
... ...
192 224
... ...
255 255

How to Generate a Look-Up Table
For example:
Let a = 2.
Let x {0,…,255}
(x ; a )=
255
1 + e a ( x 127 ) / 32
Or in Matlab:
a = 2;
x = 0:255; This is just
LUT = 255 ./ (1+exp(-a*(x-127)/32));
one example.

Point Processes: Increase Brightness

R+G+ B
I=
3
255
127
I k (r, c )+ g , if I k (r, c )+ g < 256
J k (r, c )=
g
255, if I k (r, c )+ g > 255
0
g 0 and k {1,2 ,3 } is the band index.
0 127 255
transform mapping

Point Processes: Decrease Brightness
255
255-g
127
0, if I k (r, c ) g < 0
J k (r, c )=
I k (r, c ) g , if I k (r, c )
0
g 0 and k {1,2 ,3 } is the band index.
0 127 255
transform mapping

Point Processes: Increase Contrast
255
127
Let Tk ( r, c ) = a [I k (r, c ) 127]+ 127, where a > 1.0
0, if Tk (r, c ) < 0,

J k (r, c ) = Tk (r, c ), if 0 Tk (r, c ) 255,
0
0 127 255
255, if Tk (r, c ) > 255. k {1,2 ,3 } transform mapping


Point Processes: Decrease Contrast
255
127
Tk ( r, c ) = a[I k (r, c ) 127]+ 127,
where 0 a < 1.0 and k {1, 2, 3 }.

0
0 127 255
transform mapping

Point Processes: Contrast Stretch
255
MJ
127
Let mI = min[I (r, c )], M I = max[I (r, c )],
mJ = min[J (r, c )], M J = max[J (r, c )].
Then,
I (r, c ) mI mJ
0
J (r, c )= (M J mJ ) + mJ . 0 mI 127 MI 255
M I mI transform mapping

Information Loss from Contrast

Adjustment
orig
lo-c
hi-c
histograms
Information Loss from Contrast

Adjustment
orig lo-c hi-c abbreviations:
original
low-contrast
high-contrast
restored
difference
orig lo-c lo-c
difference between
original and restored
rest diff low-contrast
orig hi-c hi-c

difference between
original and restored
rest diff high-contrast

Point Processes: Increased Gamma
255
127
1/
I (r, c )
J (r, c ) = 255 for > 1.0
255
0
0 127 255
transform mapping

Point Processes: Decreased Gamma
255
127
1/
I (r, c )
J (r, c ) = 255 for < 1.0
255
0
0 m 127 M 255
transform mapping

Gamma Correction: Effect on

Histogram

The Probability Density Function of an Image
255
Let A = hI k (g + 1).
g =0
Note that since hI k (g + 1)is the number of pixels in

I k (the k th color band of image I ) with value g,
A is the number of pixels in I . That is if I is
R rows by C columns then A = R C.
Then, This is the probability

pdf 1 that an arbitrary pixel
pI k (g + 1) = hI k (g + 1)
[lower case] A from Ik has value g.
is the graylevel probability density function of I k .

The Probability Density Function of an Image
• pband(g+1) is the fraction of pixels in (a specific band of) an

image that have intensity value g.
• pband(g+1) is the probability that a pixel randomly selected
from the given band has intensity value g.
• Whereas the sum of the histogram hband(g+1) over all g from
1 to 256 is equal to the number of pixels in the image, the
sum of pband(g+1) over all g is 1.
• pband is the normalized histogram of the band.

The Probability Distribution Function of an

Image
Let q = [q1 q2 q3] = I(r,c) be the value of a PDF

randomly selected pixel from I. Let g be a
[upper case]
specific gray level. The probability that qk
g is given by
g
g g h ( + 1)
Ik
PI k (g + 1) = pI k ( + 1) = hI k ( + 1) =
1 =0
255 ,
h ( + 1)
=0 A =0
Ik
=0
where hIk( +1) is

the histogram of
This is the probability that
the kth band of I. any given pixel from Ik has
value less than or equal to g.


Image
Let q = [q1 q2 q3] = I(r,c) be the value of a Also called CDF

randomly selected pixel from I. Let g be a for “Cumulative
specific gray level. The probability that qk Distribution
g is given by
Function”.
g
g g h ( + 1)
Ik
PI k (g + 1) = pI k ( + 1) = hI k ( + 1) =
1 =0
255 ,
h ( + 1)
=0 A =0
Ik
=0
where hIk( +1) is

the histogram of
This is the probability that
the kth band of I. any given pixel from Ik has
value less than or equal to g.


Image
A.k.a. Cumulative Distribution Function.
• Pband(g+1) is the fraction of pixels in (a specific band of) an

image that have intensity values less than or equal to g.
• Pband(g+1) is the probability that a pixel randomly selected from
the given band has an intensity value less than or equal to g.
• Pband(g+1) is the cumulative (or running) sum of pband(g+1)
from 0 through g inclusive.
• Pband(1) = pband(1) and Pband(256) = 1; Pband(g+1) is non-
decreasing.
Note: the Probability Distribution Function (PDF, capital letters) and the Cumulative Distribution Function (CDF)
are exactly the same things. Both PDF and CDF will refer to it. However, pdf (small letters) is the density function.

Intensity Transform Requirements
a) A monotonically increasing intensity transformation prevents an

intensity reversal which could cause artifacts in the transformed
image.
b) A strictly monotonically increasing intensity transformation
guarantees that the inverse transformation (from s=T(r) back to
r) will be 1:1 preventing ambiguities

Uniform Histogram Transform
These are pdf’s.

Gonzalez uses PDF
instead of pdf.
s = T ( r ) = ( L 1) pr ( w ) dw L is the number of
gray levels.
0
Cumulative Distribution Function (CDF)

Histogram Equalization Example
Note: this same 3-bit (8 gray level) 64x64 pixel image will
be used for several examples

Histogram Equalization Example
( )
0
pdf s0 = T ( r0 ) = ( 8 1) pr rj = 7 pr ( r0 ) = 7 ( 0.19 ) = 1.33

= T ( r ) = ( 8 1) p ( r ) = 7 p ( r ) + 7 p ( r ) = 7 ( 0.19 ) + 7 ( 0.25 ) = 1.33 + 1.75 = 3.08
j1= 0
s1 1 r j r 0 r 1
j=0
= T ( r ) = ( 8 1) p ( r ) = 7 ( 0.19 ) + 7 ( 0.25 ) + 7 ( 0.21) = 1.33 + 1.75 + 1.47 = 4.55

2
s2 2 r j
j=0
s3 = 5.67 s4 = 6.23 s5 = 6.65 s6 = 6.86 s7 = 7.00
Discrete CDF (transformation)

Input Output Discrete Output
0 1.33 1
1 3.08 3
2 4.55 5
3 5.67 6
4 6.23 6
Computed PDF 5 6.65 7
6 6.86 7
7 7.00 7
© 2002 R. C. Gonzalez & R. E. Woods 1999-2007 by Richard Alan Peters II

Histogram Equalization
Resulting histogram
when trying to achieve
uniform* equalized
histogram
*Resulting histogram
( )
k
sk = T ( rk ) = ( L 1) pr rj
j=0 is usually not uniform
due to discrete
Must round off to nature of transform.
nearest integer value In this case we only
since only L integer
levels in output have five output
levels.

dark
light
Low contrast
high contrast

( )
k
sk = T ( rk ) = ( L 1) pr rj
j=0
Histogram Specification
G(z) would be a
straight line for
equalization.
3. Compute CDF vq for desired histogram
1. Compute pdf (histogram) for input

image Inverting a discrete function is
far easier than inverting a
2. Compute CDF sk for this PDF continuous function
4. Match CDF’s for each sk to

create mapping table by
inverting G
Specified pdf
Original pdf
Actual transformed
CDF of specified pdf
image pdf
Histogram Specification Example

1. Compute the PDF of the image to be transformed
( )
0
s0 = T ( r0 ) = ( 8 1) pr rj = 7 pr ( r0 ) = 7 ( 0.19 ) = 1.33
= T ( r ) = ( 8 1) p ( r ) = 7 p ( r ) + 7 p ( r ) = 7 ( 0.19 ) + 7 ( 0.25 ) = 1.33 + 1.75 = 3.08
j1= 0
s1 1 r j r 0 r 1
j=0
= T ( r ) = ( 8 1) p ( r ) = 7 ( 0.19 ) + 7 ( 0.25 ) + 7 ( 0.21) = 1.33 + 1.75 + 1.47 = 4.55

2
s2 2 r j
j=0
s3 = 5.67 s4 = 6.23 s5 = 6.65 s6 = 6.86 s7 = 7.00
Discrete CDF of input histogram

0 1.33 1 Rounded off
1 3.08 3
2 4.55 5
3 5.67 6
4 6.23 6
Output CDF 5 6.65 7
(not discrete) 6 6.86 7
7 7.00 7


2. Compute the PDF for the specified histogram
( )
0
G ( z0 ) = ( 8 1) pz z j = 7 pz ( z0 ) = 7 ( 0 ) = 0
G ( z ) = ( 8 1) p ( z ) = 7 p ( z ) + 7 p ( z ) = 7 ( 0 ) + 7 ( 0 ) = 0
j1= 0
1 z j z 0 z 1
j=0
G ( z ) = ( 8 1) p ( z ) = 7 ( 0 ) + 7 ( 0 ) + 7 ( 0 ) = 0
2
2 z j
j=0
G ( z3 ) = 1.05 G ( z4 ) = 2.45 G ( z5 ) = 4.55 G ( z6 ) = 5.95 G ( z7 ) = 7.00
Discrete CDF of desired histogram

0 0 0 Rounded off
1 0 0
2 0 0
3 1.05 1
4 2.45 2
5 4.55 5
6 5.95 6
7 7.00 7


3. Match the PDFs to get the transformation
Discrete CDF desired

Discrete CDF of input histogram Pixel CDF Discrete CDF
Input CDF Discrete CDF 0 0 0
0 1.33 1 1 0 0
1 3.08 3 2 0 0
2 4.55 5 3 1.05 1
3 5.67 6 4 2.45 2
4 6.23 6 5 4.55 5
5 6.65 7 6 5.95 6
6 6.86 7 7 7.00 7
7 7.00 7
Discrete CDF of input histogram

Input CDF input CDF desired New Value
0 1 (1.33) 1 (1.05) 3
1 3 (3.08) 2 (2.45) 4
2 5 (4.55) 5 (4.55) 5
3 6 (5.67) 6 (5.95) 6
4 6 (6.23) 6 (5.95) 6
5 7 (6.65) 7 (7.00) 7
6 7 (6.86) 7 (7.00) 7
7 7 (7.00) 7 (7.00) 7

4. Put mapping in a LUT
Input Transformed Value

0 --> 3
1 --> 4
2 --> 5
3 --> 6
4 --> 6
5 --> 7
6 --> 7
7 --> 7
MATLAB: Histogram Equalization
HISTOGRAM EQUALIZATION IN MATLAB
>> f=imread(‘fig10.10(a).jpg’); % load in figure 3.15(a)

>> imshow(f) % show image in a window
>> figure, imhist(f) % show histogram in a new window
% This histogram is not normalized.
>> ylim(‘auto’) % set histogram tick marks and
% axis limits automatically
>> g=histeq(f,256); % Create histogram equalized image
% you can also do this with
% the cumsum function
>> hnorm=imhist(f)./numel(f); % compute normalized histogram
>> figure, imshow(g) % show this figure in a new window
>> figure, imhist(g) % generate a histogram of
% the equalized image
>> ylim(‘auto’) % set limits again
SEE GWE, Section 3.3.3 for a discussion of histogram specification using MATLAB
MATLAB: Useful Transformations
THRESHOLDING
>> f=imread(‘fig10.10(a).jpg’); % load in an image

>> g=(f>100)*255 % show figure in a new window
(f>100) evaluates to 1 (true) if the pixel is >100, or 0 (false) if the pixel is 100
Multiplying by 255 is necessary to have a uint8 255 level image.
ROTATION
>> f=imread(‘fig10.10(a).jpg’); % load in an image

>> g=imrotate(f,angle,’method’); % rotate it by angle
%Supports methods: ‘nearest’, ‘bilinear’, ’bicubic’

bilinear
MATLAB: Getting Picture Coordinates
INPUT CURSOR VALUES

%(Very useful for specifying reference points in an image)
>> [x,y]=ginput
%Displays the graph window, displays a cross hair. Will input
%coordinate pairs from the graph window until you type return.
%Position cursor using mouse and click to input each coordinate pair.
x=
778.3362
619.7249
844.4242
465.5196
y=
213.0831
327.6357
470.8264
609.6112
>> [x,y]=ginput(n)
Will input n coordinate pairs until you press return
Examples of Point Processing
- gamma - brightness original + brightness + gamma
histogram mod - contrast original + contrast histogram EQ

Histogram
Manually
specified
histogram h
(1) CDF sk=G(zq)

corresponding to h
(2) Inverse transform z=G-1(sk)
for intensity mapping
Resulting
histogram
Local Histogram Operations

Local Histogram Operations
Ei f ( x, y ) if mSxy k0 mg AND k1 G Sxy k2 G

g ( x, y ) =
f ( x, y ) otherwise
compare to global
mean mG and standard
deviation G
Spatial Neighborhoods

Spatial Filtering
Let I and J be images such that J = T [I].

T [·] represents a transformation, such that,
J (r, c ) = T [I ](r, c ) =
f ({I (u, v ) u {r s,..., r,...r + s}, v {c d ,..., c,...c + d }}).
That is, the value of the transformed image, J, at pixel location

(r,c) is a function of the values of the original image, I, in a 2s+1
2d+1 rectangular neighborhood centered on pixel location (r,c).

Moving Windows
• The value, J(r,c) = T[I](r,c), is a function of a rectangular

neighborhood centered on pixel location (r,c) in I.
• There is a different neighborhood for each pixel location,
but if the dimensions of the neighbor-hood are the same for
each location, then trans-form T is sometimes called a
moving window transform.

Spatial Filtering

Moving-Window Transformations
Neutral
Buoyancy We’ll take a
Facility at section of
NASA this image to
Johnson demonstrate
Space
the MWT
Center
photo: R.A.Peters II, 1999

Lecture5 PDF

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

Lecture5 PDF

Caricato da

Copyright:

Formati disponibili

EECS490: Digital Image Processing

• Point transformations (cont.)

Brightness & Contrast

© 2002 R. C. Gonzalez & R. E. Woods

Point Ops via Functional Mappings

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

One-band Image J (r, c )= f (I (r, c )),

Three-band Image J (r, c, b)= f (I (r, c, b)), or

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

One-band Image J (r, c )= f (I (r, c )),

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

A look-up table (LUT) … then the LUT

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

E.g.: index value

0 127 255 ... ...

1999-2007 by Richard Alan Peters II

Point Ops via Functional Mappings

1999-2007 by Richard Alan Peters II

How to Generate a Look-Up Table

1999-2007 by Richard Alan Peters II

Point Processes: Increase Brightness

1999-2007 by Richard Alan Peters II

Point Processes: Decrease Brightness

1999-2007 by Richard Alan Peters II

Point Processes: Increase Contrast

1999-2007 by Richard Alan Peters II

Point Processes: Decrease Contrast

where 0 a < 1.0 and k {1, 2, 3 }.

1999-2007 by Richard Alan Peters II

Point Processes: Contrast Stretch

1999-2007 by Richard Alan Peters II

Information Loss from Contrast

Information Loss from Contrast

orig hi-c hi-c

1999-2007 by Richard Alan Peters II

Point Processes: Increased Gamma

1999-2007 by Richard Alan Peters II

Point Processes: Decreased Gamma

1999-2007 by Richard Alan Peters II

Gamma Correction: Effect on

1999-2007 by Richard Alan Peters II

The Probability Density Function of an Image

Note that since hI k (g + 1)is the number of pixels in

Then, This is the probability

1999-2007 by Richard Alan Peters II

The Probability Density Function of an Image

• pband(g+1) is the fraction of pixels in (a specific band of) an

1999-2007 by Richard Alan Peters II

The Probability Distribution Function of an

Let q = [q1 q2 q3] = I(r,c) be the value of a PDF

where hIk( +1) is

1999-2007 by Richard Alan Peters II

The Probability Distribution Function of an

Let q = [q1 q2 q3] = I(r,c) be the value of a Also called CDF

where hIk( +1) is

1999-2007 by Richard Alan Peters II

The Probability Distribution Function of an

Ei f ( x, y ) if mSxy k0 mg AND k1 G Sxy k2 G