IMAGE FUNDAMENTALS
A Simple Image Model
Image refers to a two dimensional light intensity function
denoted by f(x,y) - x and y denote the spatial co-ordinates.
The magnitude of f( ) at spatial co-ordinates (x,y) gives the
intensity of image at that point.
f(x,y) can be characterized by two components.
The amount of source light being incident on the scene being viewed –
illumination component (i(x,y)).
The amount of light reflected by the objects in the scene – reflectance
component (r(x,y)).
Cont..
f(x,y) = i(x,y) r(x,y)
0 < i(x,y) < ∞
0 < r(x,y) < 1
f(x,y) can be treated as an MxN array for
processing, where M and N are the dimension of
the image along X and Y directions.
Each element of the array represents the intensity
values and can be called as pixel or pel.
Image Resolution
Resolution:
Image Resolution = the dpi (dots per inch = pixels per
inch) of an image.
Cont..
Common image resolutions are 72, 200, 300, etc.
The higher the resolution, the higher the print
quality and the larger the data size of the image (in
kilobytes or megabytes)
The resolution of a digital image is defined as the
number of pixels it contains. A 5 megapixel image is
typically 2,560 pixels wide and 1,920 pixels high
and has a resolution of 4,915,200 pixels, rounded off
to 5 million pixels.
2-D Function Representation
2-D Function Representation
2
2 3 2
2
2D function – Examples.
n2
A unit Step sequence u(n1, n2)
n1
n2
A unit impulse δ(n1, n2)
n1
2D function – Examples.
n2
Line impulse δT(n1)
n1
Separable Functions
A 2-D sequence x(n1,n2) is said to be a
separable sequence if it can be expressed as
x(n1, n2) = f(n1) g(n2)
δ (n1, n2) = δ(n1) δ(n2).
U(n1, n2) = u(n1) u(n2).
Two Dimensional Systems
Linear System:
Let x(m,n) and y(m,n) represent the input and output of a
two dimensional system,
y(m,n) = T [x(m,n)]
The system is linear iff any linear combination of two
inputs x1(m,n) and x2(m,n) produces the combination of
their respective outputs y1(m,n) and y2(m,n) for arbitrary
constants a1 and a2.
T[a1 x1(m,n) + a2 x2(m,n)] = a1T[x1(m,n)] + a2T[ x2(m,n)]
= a1 y1(m,n) + a2 y2(m,n)
Cont.
Shift Invariant System:
The system is spatially invariant or shift invariant
if a translation of input causes the same
translation of the output.
If y(m,n) = T [x(m,n)]
y(m-i,n-j) = T [x(m-i,n-j)]
2-D Convolution
The 2D discrete convolution equation is
M 1 N 1
C (i, j ) A(k , l ) B(i k , j l )
k 0 l 0
Steps for 2-D convolution
Either of the array is rotated by 180 degree. (say B
(k, l))
Then shifted by (i, j) and overlayed on the other
array (A (k, l))
Sum the product of arrays in the overlapping
regions gives the result at (i, j).
Convolution - Illustration
2-D Convolution - Example
Matrix Method
If F and G are the input arrays of size m1 x n1 and
m2 x n2 resp.
Pad both the matrices with zeros to bring them to a
size M x N
M ≥ m1 + m2 -1
N ≥ n1 + n2 -1
Form an MN x 1 column vector fp from matrix F by
row stacking.
Form a similar matrix Gp from G matrix.
From an MN x MN circulant matrix from the rows
of Gp.
Matrix Method Cont..
The circulant matrix formed is called a block
circulant matrix, Gb, which can be written as
[G1 ][G N ]........[G2 ]
[G
2 1][G ].........[G ]
3
Gb . . .
. . .
[G ][G ]....[G ]
N N 1 1
Each N x N block matrix is formed from the ith row
of Gp
Example – 2D Convolution
1 2 0 1 1 0
1 2 1 1
F G f p 3 4 0 g p 2 2 0
3 4 2 2 0 0 0 0 0 0
1 0 1 0 0 0 2 0 2 1
1 1 0 0 0 0 2 2 0 2
0 1 1 0 0 0 0 2 2 0
2 0 2 1 0 1 0 0 0 3
Gb 2 2 0 0 Y = G b * fp ’
1 1 0 0 0 f p ' 4
0 2 2 0 1 1 0 0 0 0
0 0 0 2 0 2 1 0 1 0
0 0 0 2 2 0 1 1 0 0
0 0 0 0 2 2 0 1 1 0
Example – 2D Convolution
1
1 0 0 0 0 2 2 0 1 2
1 1 0
0 0 0 2 2 0 0
0 1 1 0 0 0 0 2 2
3
2 0 1 1 0 0 0 0 2 f p ' 4
Gb ' 2 2 0 1 1 0 0 0 0
0
0 2 2 0 1 1 0 0 0 0
0
0 2 2 0 1 1 0 0 0
0 0 0 2 2 0 1 1 0
0
0 0 0 0 2 2 0 1 1
Y = Gb ’ * fp ’
Circular convolution
If two 1-D convolving sequences are periodic then
their convolution is also periodic and can be
represented as
N 1
y ( n) x ( k ) h( n k ) 0 n N 1
k 0
h(n)=h(N-n) and N-period
Cont..
Example:
N=4, h(n) = n+3(modulo 4)
h(n) = [3 0 1 2] x(n) = [x(0) x(1) x(2) x(3)]
y(0) = 3*x(0) + 2*(x(1) + 1*x(2) + 0*x(3)
y(1) = 0*x(0) + 3*(x(1) + 2*x(2) + 1*x(3)
y(2) = 1*x(0) + 0*(x(1) + 3*x(2) + 2*x(3)
y(3) = 2*x(0) + 1*(x(1) + 0*x(2) + 3*x(3)
Cont..
y ( 0) 3 2 1 0 x ( 0)
y (1) 0 3 2 1 x(1)
y ( n) H c x ( n)
y ( 2) 1 0 3 2 x ( 2)
y ( 3) 2 1 0 3 x(3)
Hc, can be obtained by arranging the row vector h(n)
column wise and then obtaining the remaining columns by
rotating the previous column by one.
Circular convolution – 2D
The circular convolution in 2-D can be
implemented in using the same principles
using a block circulant matrix.
Separable Function an eg.
1 1
h(n1 , n2 )
1 1
1
h2 (n2 ) 1 1 h1 (n1 )
1
1 1 1
1 1 1 1 1
h(0,0) = h1(0) x h2(0) = 1.1 = 1; h(0,1) = h1(1) x h2(0) = 1.-1 = -1
h(1,0) = h1(0) x h2(1) = 1.1 = 1; h(1,1) = h1(1) x h2(1) = 1.-1 = -1
2-D Correlation
The 2D discrete correlation equation is
M 1 N 1
C (i, j ) A(k , l ) B (i k , j l )
k 0 l 0
