Professional Documents
Culture Documents
Thi-Giac-May-Tinh - Vo-Quang-Hoang-Khang - Dip02-Image-Enhancement-Point-Processing - (Cuuduongthancong - Com)
Thi-Giac-May-Tinh - Vo-Quang-Hoang-Khang - Dip02-Image-Enhancement-Point-Processing - (Cuuduongthancong - Com)
om
.c
Image Enhancement
ng
co
Point Processing
an
th
o ng
du
u
cu
CuuDuongThanCong.com https://fb.com/tailieudientucntt
Contents
om
processing techniques:
.c
What is point processing?
ng
Negative images
co
Thresholding
an
Logarithmic transformation
Power law transforms th
o ng
Grey level slicing
du
om
reduced to the form
Origin y
.c
g (x, y) = T[ f (x, y)]
ng
where f (x, y) is the
co
input image, g (x, y) is
an
the processed image
th
and T is some ng (x, y)
operator defined over
o
du
some neighbourhood
u
of (x, y)
cu
x Image f (x, y)
om
neighbourhood is simply the pixel itself
.c
In this case T is referred to as a grey level
ng
transformation function or a point processing
co
operation.
an
Point processing operations take the form
s=T(r) th
o ng
where s refers to the processed image pixel value and r
du
om
grey detail embedded in dark regions of an image
.c
Note how much clearer the tissue is in the negative
ng
image of the mammogram below
co
an
th
o ng
Original Negative
s = intensitymax- r
du
Image Image
u
cu
om
Original Image Enhanced Image y
y
.c
ng
co
an
th
ng
x Image f (x, y) x Image f (x, y)
o
du
u
cu
s = intensitymax - r
om
segmentation in which we want to isolate an object of
.c
interest from a background.
ng
co
an
th
1.0 r > threshold
ng
s=
o
0.0 r <= threshold
du
u
cu
om
.c
ng
co
an
th
ng x
x Image f (x, y) Image f (x, y)
o
du
s=
0.0 r <= threshold
om
.c
ng
co
an
th
o ng
du
u
cu
om
transformations
.c
Three of the most
ng
common are shown
co
here
an
• Linear
th
- Negative/Identity ng
• Logarithmic
o
du
- Log/Inverse log
• Power law
u
cu
om
s = c * log(1 + r)
.c
The log transformation maps a narrow range of low grey
ng
level values in the input image into a wider range of output
co
values. The opposite is true of higher values of the input
an
levels.
th
We use the log transform to expand the values of the dark
ng
pixels in an image while compressing the higher level
o
values.
du
om
grey level values may have an extremely large range
.c
of values.
In the following example the Fourier transform of an
ng
co
image is put through a log transform to reveal more
an
detail
th
o ng
du
s = log(1 + r)
u
cu
om
.c
ng
co
an
th
ng
x Image f (x, y) x Image f (x, y)
o
du
s = log(1 + r)
u
We usually set c to 1
cu
om
s=c*rγ
.c
Map a narrow range
ng
of dark input values
co
into a wider range of
an
output values, with
th
the opposite being true ng
for the higher values
o
du
u
family of curves
om
.c
ng
co
an
th
ng
y Image f (x, y) y Image f (x, y)
o
du
s=rγ
u
We usually set c to 1
cu
om
.c
ng
co
an
th
o ng
du
u
cu
γ = 0.6
om
1
.c
Transformed Intensities
0.9
0.8
ng
0.7
co
0.6
0.5
an
0.4
th
0.3 ng
0.2
0.1
o
0
du
Old Intensities
cu
γ = 0.4
om
1
.c
0.9
Transformed Intensities
0.8
ng
0.7
co
0.6
0.5
an
0.4
0.3
th
0.2 ng
0.1
0
o
du
γ = 0.3
om
1
.c
0.9
Transformed Intensities
0.8
ng
0.7
co
0.6
0.5
an
0.4
0.3
th
0.2 ng
0.1
0
o
du
om
right show a s = r 0.6
.c
magnetic resonance
(MR) image of a
ng
co
fractured human
s = r 0.4
spine
an
th
Different curves ng
highlight different
o
detail
du
u
cu
γ = 5.0
om
.c
1
0.9
Transformed Intensities
ng
0.8
0.7
co
0.6
0.5
an
0.4
0.3
th
0.2 ng
0.1
0
o
0 0.2 0.4 0.6 0.8 1
du
Original Intensities
u
cu
An aerial photo
of a runway is
om
shown s = r 3.0
.c
This time
ng
power law
co
transforms are
s = r 4.0
an
used to darken
th
the image ng
Different curves
o
du
highlight
different detail
u
cu
om
of computer monitors.
.c
Problem is that
ng
display devices do
co
not respond linearly
an
to different
th
intensities ng
Can be corrected
o
du
using a log
u
transform
cu
om
.c
ng
co
an
th
o ng
du
u
cu
om
The images below show a contrast stretching linear
.c
transform to add contrast to a poor quality image
ng
co
an
th
o ng
du
u
cu
om
.c
ng
co
an
th
o ng
du
u
cu
om
.c
ng
co
an
th
o ng
du
u
cu
om
.c
ng
co
an
th
o ng
du
u
cu
om
• Similar to thresholding
.c
• Other levels can be
ng
suppressed or maintained
co
• Useful for highlighting features
an
in an image
th
o ng
du
u
cu
om
an image we can highlight interesting aspects of that
.c
image
• Higher-order bits usually contain most of the significant
ng
co
visual information
• Lower-order bits contain
an
th
subtle details ng
o
du
u
cu
om
[10000000] [01000000]
.c
ng
co
an
[00100000] [00001000]
th
o ng
du
u
cu
[00000100] [00000001]
om
.c
ng
co
an
th
o ng
du
u
cu
Reconstructed image
om
using only bit planes 8
.c
and 7
ng
co
Reconstructed image
an
using only bit planes 8, 7
th and 6
o ng
du
Reconstructed image
u
cu
om
image enhancement
.c
Next time we will start to look at image enhancement-
ng
histogram equalization
co
an
th
o ng
du
u
cu