Histogram Specification

desired
­1
equalize G equalize G
 Histogram Specification (cont.)
● Equalize the levels of the original image.
● Specify the desired density function
(histogram) and obtain the transformation
function G(z).
● Apply the inverse transformation function
-1
z = G (s) to the levels obtained in first step.
 Histogram Specification (cont.)
­from text­
● Saw what we want the histogram to look like and
come up with a transform function that will give
it to us.
● Continuous random variables r & z. Pr(r) and
Pz(z) denote their probability density functions.
(Continuous equivalent of a histogram).
● Pr(r) input image
● Pz(z) desired output image (processed) This is
what we'd like the processed image to have.
 Histogram Specification (cont.)
­from text­
● Continuous version of histogram equalization.
● Cumulative Distribution Function

s =T r = ∫ Pr w dw
0
 Histogram Specification (cont.)
­from text­
● Continuous version of histogram equalization.
● Cumulative distribution function.
r

s=T r =∫ P r  wdw

∫ P r  wdw
0
Transformation G(z) can be obtained by
using z
since you
∫ P z t dt specified Pz(z)!
0
 Histogram Specification (cont.)
­from text­
-1
● Assume G exists and that is satisfys:
● Single­Valued, monotonically increasing
-1
● 0 <= G (z) <= 1 for 0 <= z <= 1
 Histogram Specification (cont.)
­from text­
● You can obtain an image with the specified
probability density function from an input
function using the following:
– obtain T(r) using equilization
– obtain G(z) by equalizing specified histogram
– obtain G (z)
-1
– obtain the output image by applying G to all-1pixels
in input image.
 Histogram Specification (cont.)
­from text­
● So far so good (continuously speaking)
● In practice it is difficult to obtain analytical
-1
expressions for T(r) and G
● With discrete values it becomes possible to
make a close approximation to the histogram.
discrete Formulas
k k
nj
s k =T r k = ∑ P r r j = ∑ k =0,1,2,... , L −1
j =0 j =0 n
L is the number of discrete gray levels
n = total # pixels
n = # of pixels with gray level j
j
k
v k =G  z k = ∑ p z  z i =s k k =0,1,2,. .. , L −1
i=0
−1
z k =G [T r k ] k =0,1,2,. .. , L −1
−1
z k =G  s k  k =0,1,2,. .. , L −1
 Implementation
● Each set of gray levels is a 1D array
● All mappings from r to s and s to z are simple
table lookups
● Each element (e.g. s ) contains
k 2 important
pieces of information:
– subscript k denotes the location of the element in
the array
– s denotes the value at that location
● We need to only look at integer values
 Refer to Figure 3.19 GW
● (a) a hypothetical transform function given an
image that was equalized.
-1
● (b) Equalize specified histogram. G is just
running the transform backwards.
● But wait a minute! Where did we get the z's???
● We have to use an iterative scheme.
 Iterative Scheme
● Obtain histogram of the given image.
● Equalize the image to precompute a mapped
level sk for each rk . T
● Obtain G from the specified histogram by
equalization.
● Precompute z k for each sk using iterative
scheme (3.3­17)
● Map r k­> s k and back to z k .
● Moon example 3.4, page 100, 101, 102 GW
 desired
0.3 0.3
0.2 0.2
0.1 0.1

0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
0 – 0.19 0 – 0.0
1 – 0.25 1.0 1 – 0.0
1.0
2 – 0.21 2 – 0.0
0.8 0.8
3 – 0.16 3 – 0.15
0.6 0.6
4 – 0.08 4 – 0.20
0.4 0.4
5 – 0.06 5 – 0.30
0.2 0.2
6 – 0.03 6 – 0.20
0.0 7 – 0.15 0.0
7 – 0.02
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7

7 1.0
6 0.8
5
4 0.6
3 0.4
2 0.2
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0
6 0.8
5
4 0.6
3 0.4
2 0.2
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6 0.8
5
4 0.6
3 0.4
2 0.2
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6
5
0.8 3 6
4 0.6
3 0.4
2 0.2
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6
5
0.8 3 6
0.6
4
3
4 6
0.4
2 0.2
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6
5
0.8 3 6
0.6
4
3
4 6
0.4
2 0.2 5 7
1
0 0.0
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6
5
0.8 3 6
0.6
4
3
4 6
0.4
2 0.2 5 7
1
0 0.0 6 7
0 1 2 3 4 5 6 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
1 3
7 1.0 2 5
6
5
0.8 3 6
0.6
4
3
4 6
0.4
2 0.2 5 7
1
0 0.0 6 7
0 1 2 3 4 5 6 7 7 7
 equalized histogram
0.3 0.3
0.2 0.2
0.1 0.1
r s
0 1
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
1 3
2 5 1 = 0.19
7 1.0
6 0.8 3 6 3 = 0.25
5
0.6 5 = 0.21
4
3
4 6
0.4
5 6 = .16+.08=.24
2 0.2 7
1
7 7 = .06+.03+.02
0 0.0 6
= 0.11
0 1 2 3 4 5 6 7 7 7
 desired equalized histogram

0.3
0.3
0.2
0.2
0.1 r s 0.1
0 0
0 1 2 3 4 5 6 7
0 1 2 3 4 5 6 7 1 0
2 0
7 1.0
6 3 1
0.8
5
4 0.6 4 2
3 0.4 5 4
2 0.2
1 6 6
0.0
0 7 7
0 1 2 3 4 5 6 7
 0 1 3
equalized orig resultant histogram
1 3 4 0.3 0.3
2 5 5 0.2 0.2
3 6 6
0.1 0.1
4 6 6
5 7 7
0 1 2 3 4 5 6 7
6 7 7 0 1 2 3 4 5 6 7

7 7 7
7 1.0
6 0.3
0.8
5
4 0.6 0.2
3 0.4
2 0.2
0.1
1
0 0.0
0 1 2 3 4 5 6 7
desired 0 1 2 3 4 5 6 7