Image Pyramids

CoE4TN3
Image Processing

Wavelet and Multiresolution

Processing

Introduction Image pyramids

Unlike Fourier transform, whose basis functions are
sinusoids, wavelet transforms are based on small waves,
called wavelets, of limited duration.
Fourier transform provides only frequency information,
but wavelet transform provides time-frequency
information.
Wavelets lead to a multiresolution analysis of signals.
Multiresolution analysis: representation of a signal (e.g.,
an images) in more than one resolution/scale.
Features that might go undetected at one resolution may
be easy to spot in another.
At each level we have an approximation image and a
residual image.
The original image (which is at the base of pyramid)
and its P approximation form the approximation
pyramid.
The residual outputs form the residual pyramid.
Approximation and residual pyramids are computed
in an iterative fashion.
A P+1 level pyramid is build by executing the
operations in the block diagram P times.

2 5

Multiresolution Image pyramids

During the first iteration, the original 2Jx2J image is

applied as the input image.
This produces the level J-1 approximate and level J
prediction residual results
For iterations j=J-1, J-2, , J-p+1, the previous
iterations level j-1 approximation output is used as
the input.

3 6

1
 Image pyramids Subband coding
Each iteration is composed of three sequential steps: In subband coding, an image is decomposed into a set of
bandlimited components, called subbands.
1. Compute a reduced resolution approximation of the
Since the bandwidth of the resulting subbands is smaller than
input image. This is done by filtering the input and
that of the original image, the subbands can be downsampled
downsampling (subsampling) the filtered result by a without loss of information.
factor of 2.
Filter: neighborhood averaging, Gaussian
filtering
The quality of the generated approximation is a
function of the filter selected

7 10

Image pyramids Perfect Reconstruction Filter

2. Upsample output of the previous step by a factor of
2 and filter the result. This creates a prediction
image with the same resolution as the input.
By interpolating intensities between the pixels of
step 1, the interpolation filter determines how
accurately
t l the
th prediction
di ti approximates
i t theth input
i t Z transform: X ( z ) = 1 [ H 0 ( z )G0 ( z ) + H1 ( z )G1 ( z ) ] X ( z )
2
to step 1. 1
+ [ H 0 ( z )G0 ( z ) + H1 ( z )G1 ( z ) ] X ( z )
3. Compute the difference between the prediction of 2
step 2 and the input to step 1. This difference can be Goal: find H0, H1, G0 and G1 so that

( )
later used to reconstruct progressively the original
x(n) = x (n) i.e. X ( z ) = X ( z )
image

8 11

Perfect Reconstruction Filter: Conditions

H 0 ( z )G0 ( z ) + H1 ( z )G1 ( z ) = 0
If
H 0 ( z )G0 ( z ) + H1 ( z )G1 ( z ) = 2

Then
X ( z ) = X ( z )

9 12

2
 Perfect Reconstruction Filter Families

QMF: quadrature mirror filters

CQF: conjugate mirror filters

13 16

2-D The Haar Transform

Haar proposed the Haar Transform in 1910, more
than 70 years before the wavelet theory was born.
Actually, Haar Transform employs the Haar wavelet
filters but is expressed in a matrix form.
Haar wavelet is the oldest and simplest wavelet basis.
basis
Haar wavelet is the only one wavelet basis, which
holds the properties of orthogonal, (anti-)symmetric
and compactly supported.

14 17

Example of Filters The Haar Wavelet Filters

h0 = { 2 2, 2 2 } {
h1 = 2 2, 2 2 }
g0 ={ 2 2, 2 2} g0 ={ 2 2, 2 2}

15 18

3
 Multiresolution Expansions
Scaling functions
Integer translations and dyadic scalings of a
scaling function ( x)

j ,k ( x) = 2 j / 2 (2 j x k )

Express f ( x) as the combination of j ,k ( x)

0

f ( x) = k j0 ,k ( x)
k

19 22

Multiresolution Expansions
Series Expansions 0 0 x < 1
( x) =
A function can be expressed as 1 otherwise
1
0, k ( x) = 1,2 k ( x)
f ( x ) = k k ( x) 2
k
1
where + 1,2
1 2 k +1 ( x )
2
k = k ( x), f ( x) = k* ( x) f ( x)dx f ( x) = 0.51,0 ( x)
+1,1 ( x) 0.251,4 ( x)
k ( x) Dual function of k ( x)
* Complex conjugate operation

20 23

Multiresolution Expansions Multiresolution Expansions

Series Expansions Scaling functions
Orthonormal basis Dilation equation for scaling function ( x)
k ( x ) = k ( x) ( x) = h (n) 2 (2 x n)
0 jk
j ( x), k ( x) =
n

1 j=k h (n) are called

ll d scaling
li function
f ti coefficients
ffi i t
biorthogonal Example: Haar wavelet, h (0) = h (1) = 1 2
j ( x), k ( x) = 0 jk 1 1
( x) = 2 (2 x) + 2 (2 x 1)
0 jk 2 2
j ( x), k ( x) =
1 j=k
21 24

4
 Multiresolution Expansions Wavelet Transform: 1-D
Wavelet functions Wavelet series expansion
( x) = h (n) 2 (2 x n)

n f ( x) = c j0 j0 ,k ( x) + d j (k ) j ,k ( x)
h (n) are called wavelet function coefficients k j = j0 k

Translation and scaling of ( x)

where
h
j ,k ( x) = 2 j / 2 (2 j x k )
c j0 (k ) = f ( x), j0 ,k ( x) = f ( x) j0 ,k ( x)dx
condition for orthogonal wavelets
d j (k ) = f ( x), j ,k ( x) = f ( x) j ,k ( x)dx
h (n) = (1) n h (1 n)

25 28

1
y ( x) = 0,0 ( x)
3
1
0,0 ( x)
4
2
1,0 ( x)
32
3 2
1,1 ( x)
32
+......

26 29

Wavelet Transform: 1-D

Haar Wavelet Discrete Wavelet Transform
1
1 0 x < 0.5
f ( x) = W ( j0 , k ) j0 ,k ( x)
M k

( x) = 1 0.5 x < 1 1
0 elsewhere

+ W ( j, k ) j ,k ( x)
M j = j0 k
where
1
Approximation
coefficients
W ( j0 , k ) =
M
f ( x)
x
j0 , k ( x)

1
f ( x)
Detail
coefficients
W ( j , k ) = j ,k ( x)
M x

27 30

5
 Fast Wavelet Transform: Decomposition Fast Wavelet Transform: Reconstruction

W ( j , k ) = h ( n) W ( j + 1, n) W ( j + 1, k ) = h (k ) Wup ( j , k ) + h (k ) Wup ( j , k ) k 0
n =2 k ,k 0

W ( j , k ) = h ( n) W ( j + 1, n) n = 2 k ,k 0

31 34

Fast Wavelet Transform: Decomposition Fast Wavelet Transform: Reconstruction

32 35

Example: Haar Wavelet Fast Wavelet Transform: Reconstruction

1 2 n=0

1/ 2 n = 0,1 h (n) = 1 2 n =1
h (n) =
0 otherwise 0 otherwise

33 36

6
Wavelet Transform vs. Fourier Transform 2-D Wavelet Transform: Reconstruction

37 40

Wavelet Transform: 2-D

Scaling function:
( x, y ) = ( x) ( y )
Wavelet functions:
H ( x, y ) = ( x) ( y ) Horizontal direction

V ( x, y ) = ( x) ( y ) Vertical direction

( x, y ) = ( x) ( y )
D
Diagonal direction

38 41

2-D Wavelet Transform: Decomposition

39 42

7
 Wavelet Transform based Denoising
Three Steps:
Fig. 7.24 (g) Decompose the image into several scales.
For each wavelet coefficient y:
y y t
Hard thresholding: y =
0 y <t
sign( y )i( y t ) y t
Soft thresholding: y =
0 y <t
Reconstruct the image with the altered
wavelet coefficients.
43 46

Image Processing by Wavelet Transform

Three Steps:
Decompose the image into wavelet domain
Alter the wavelet coefficients, according to
your applications such as denoising
denoising,
compression, edge enhancement, etc.
Reconstruct the image with the altered
wavelet coefficients.

44 47

Assignment

Get familiar with the Matlab Wavelet

Toolbox.
By using the Wavelet Toolbox functions,
write a p
program
g to realize the soft-
thresholding denoising on a noisy MRI image.

45 48

8
End of the lecture

49