Anda di halaman 1dari 33

MTI UGM

2010
 A definition of Image Processing and
Computer Vision
 Image Processing
◦ Manipulate image data
◦ Generate another image
 Computer Vision Image Processing
◦ Process image data
◦ Generate symbolic data
Computer Vision

Multimedia 16/11/2010 2
 Relationship between Image Coding, Image
Processing and Image Compression

Image Image
Coding Processing

Multimedia 16/11/2010 3
 Rectify the defects introduced by the image
capturing process
 Improve the image appearance for better
human perception
 Identify and quantify structures in the image
 Transform image into an alternative
representation in which some operations may
be performed more efficiently

Multimedia 16/11/2010 4
Three simple grey scale operations
 Add a constant to each pixel of a grey
image
 Scale the values by a constant multiplier
 Apply a linear operations
Other grey level manipulations
 False Contouring
 Thresholding
 Grey Level Histogram

Multimedia 16/11/2010 5
output
 Manipulate the grey
or colour value of
an individual pixel
 The only
information is
brightness or colour
 Map each input
grey value to an
output grey value input

16/11/2010 Multimedia 6
 Add a constant to each output
pixel of a grey image max
 The output grey values
are truncated when
they exceed the valid
range
 The entire image is
brightened up if the
constant is positive
min
 The image is darkened
if the constant is
negative
input

16/11/2010 Multimedia 7
original -10% -30%

+10% +30% +50%

Multimedia 16/11/2010 8
 Scale the values by output
a constant max

multiplier
 The difference in
brightness is
raised/decreased
when the scale is
more/less than
0
1.0
input
16/11/2010 Multimedia 9
In general, combine the two simple
operations to adjust the contrast and
overall brightness
 The brightness of the pixels in the new
image, P, can be related to the brightness
in the old image, p, by
Px, y  k  px, y  l , x, y 1, N
where the overall brightness is controlled
by a level, l, and the contrast is
controlled by a gain, k.

16/11/2010 Multimedia 10
 How much brightness should be added?
 What scaling factor ought to be used?

 One common approach: match the mean grey


value to a previously determined value

Multimedia 16/11/2010 11
Interactive manipulation
 User chooses the input grey values, a and b,
and output grey values a’ and b’.
 The system performs linear mapping from [a,
b] to [a’, b’].
 The extreme values 0 and 255 remain
unchanged.

Multimedia 16/11/2010 12
Interactive manipulation output
 Maps linearly input
grey values from [a, b] b’
to the ouptut grey
values [a’, b’]
 Certain portions [a, b]
are expanded
 Other portions are a’
compressed
a b input

16/11/2010 Multimedia 13
An automatic approach:
use the full range of output
grey level
 Find the minimum and 255
maximum brightness in
the image
 Extend the range of
min to max brightness
in picture to the full
range [0, 255] of
brightness
 A negative constant
and scale >1.0 0
increases the difference Min
in brightness Max
input

16/11/2010 Multimedia 14
+50%
-50%

-90%
Original 200x200 +90%

Multimedia 16/11/2010 15
False contouring output
 Apply the
255
sawtooth operator
to the image
colours
 Change gradual
changes into rapid
changes from dark
0
to bright values
input

16/11/2010 Multimedia 16
False Contouring
 Highlight local intensity variations
 Image divided into regions according to
pixels’ intensities

Multimedia 16/11/2010 17
 Use a threshold to transforms an input
image into a binary image
 Compare each input pixel value f(i, j)
against the threshold, θ, and setting the
output pixel value g(i, j) appropriately

0 if f (i, j )  
g (i, j )  
1 otherwise

Multimedia 16/11/2010 18
 Different images can be
generated from using
different thresholds.
 Higher threshold allows
only fewer brighter pixels
to become white.

θ=141 θ=128 θ=115

16/11/2010 Multimedia 19
 Sample black and
white images

θ=141 θ=128 θ=115

16/11/2010 Multimedia 20
 Many applications in image processing
 Difficult to decide the value of threshold, θ.
 The threshold value may depend on the
application required.

Multimedia 16/11/2010 21
The Grey Level Histogram or Intensity
Histogram
 Plots the number of pixels with a particular
brightness level against the brightness level.
 List of grey level values at the x-axis
 Number of occurrence with the grey value at
the y-axis

Multimedia 16/11/2010 22
Intensity Histogram
800
700
600
500
400
300
Lena 256x256 200
100
Intensity concentrates 0
around 5 peaks
GreyLevel

16/11/2010 Multimedia 23
Intensity Histogram
500

400

300

200

flower 256x192 100


Intensity concentrates
around 2 peaks 0
GreyLevel

Multimedia 16/11/2010 24
Intensity Histogram
12000

10000

8000

6000

4000

2000

Ratbrain MRI image 256x256 0


GreyLevel
Intensity concentrates
Near the black due to
The background

Multimedia 16/11/2010 25
 The intensity histogram shows
◦ Intensity concentrates around peaks
◦ Not all available grey levels are used.
 The image may be normalised to cover the full
range of grey levels.
 The old image O has histogram from O1 to O2.
The new image N will have histogram from N1 to
N2. The intensity of a pixel Nx,y is given by

N 2  N1
N x, y   (Ox, y  O1 )  N1 , x, y
O2  O1

16/11/2010 Multimedia 26
 Histogram equalisation
◦ Assume that the image’s appearance will be
improved if the distribution of pixels over the
available grey levels is even.
◦ The cumulative histogram is generated from
the grey level histogram
◦ The cumulative histogram, G(i), is derived
from the grey level histogram, g(i), as

i
G (i)   g (k )
k 0

16/11/2010 Multimedia 27
 The cumulative intensity histogram of the
flower image
Cumulative Intensity Histogram

60000

50000

40000

30000
20000

10000

0
0 32 64 96 128 160 192 224

Multimedia 16/11/2010 28
 The ideal case
◦ An equal number of
pixels having each
grey level
 Equalisation
Transform expand expand
◦ warps the horizontal
axis of the histogram,
◦ Expand it in regions
where the number of
pixels with a grey Ideal
value is more than the case
ideal case
◦ Compress it in other
regions where the
number of pixels is
fewer than the ideal
case compress compress compress

Multimedia 16/11/2010 29
 Maps each input, i, grey values to output, j,
grey values
 The jth entry of the cumulative histogram
should have the value j*N/l. Thus,
j*N/l=G(i)
=> j=(l/N)*G(i)
 The histogram manipulation is simple, but
it is not the best method to improve the
image’s appearance.

Multimedia 16/11/2010 30
 The program may have the following codes
For Y = 1 To maxY
For X = 1 To maxX
oldColor = Pixel(X, Y)
newColor = Int(CumulativePixels(oldColor) * (maxGrey -
minGrey) / PixelCount) Mod 256
Next X
Next Y

‘ Where CumulativePixels(oldColor) is an array that stores


the number of pixels with intensity less than or equal to
the index.

Multimedia Coding and Processing Dr. Philip Tse 31


Equalized image Slightly
darker

Multimedia Coding and Processing Dr. Philip Tse 32


Equalized image brighter

Multimedia Coding and Processing Dr. Philip Tse 33

Anda mungkin juga menyukai