Scribd
Upload
42 views

Wavelets: Going Beyond The Fourier Transform

Wavelets go beyond the Fourier transform by using localized basis functions rather than infinite sine and cosine waves. The 1D Haar wavelet represents a function using piecewise constant basis functions. The 2D Haar wavelet transform is computed by applying the 1D transform to columns and then rows. Implementing the Haar wavelet transform on images leads to compression, as the entropy of the transformed image is lower than the original image entropy.

Uploaded by

Aditya Dusi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
42 views

Wavelets: Going Beyond The Fourier Transform

Wavelets go beyond the Fourier transform by using localized basis functions rather than infinite sine and cosine waves. The 1D Haar wavelet represents a function using piecewise constant basis functions. The 2D Haar wavelet transform is computed by applying the 1D transform to columns and then rows. Implementing the Haar wavelet transform on images leads to compression, as the entropy of the transformed image is lower than the original image entropy.

Uploaded by

Aditya Dusi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 15

Wavelets

Going beyond the Fourier


transform
Recap of the Fourier Transform:
Uses Sine and cosine waves as basis
functions.
Complex quantity: Modulus (magnitude)
and phase.
Sine and cosine waves have infinite
extent while digital images have a finite
extent.
Digital Images can be represented more
efficiently by localized functions.

Haar wavelets (introduced by Haar in


1910) are the simplest wavelets. Unlike the
1-D Haar Wavelet:
Representation of a 1-D function in
terms of a set of localized basis functions.

What are these functions?

The first basis function is simply a constant


over the whole domain (0,1)
The second basis function is:
x(x)
1
0 1
-1
1 x(x)

-1

x-1(x)

-1
00 ( x)
10 (2 x)
11 (2 x 1)
20 (4 x)
21 (4 x 1)
22 (4 x 2)
23 (4 x 3)
We use these basis functions to compute
the Haar transform
Represent the discrete 1-D function as a
column vector.
The 1-D function is defined at N points

c Hf
(where N is a power of 2), then the Haar
wavelet transform is defined by:

Here the rows of H are the basis functions


defined earlier.
If N = 2 then H is given by:
1 1
H2
1 1
1 1 1 1
1
1 1 1
H4
1 1 0 0

0 0 1 1
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1

1 1 1 1 0 0 0 0

0 0 0 0 1 1 1 1
H8
1 1 0 0 0 0 0 0

0 0 1 1 0 0 0 0
0 0 0 0 1 1 0 0

0 0 0 0 0 0 1 1
What is the significance of multiplying f
with H?

The rows of H are the basis functions.


If f is written as a column vector,
multiplying f with H is equivalent to taking
the dot product of f with each of the basis
functions.
This gives us the projection of f along
each of the basis functions.
Notice that the basis functions are
orthogonal to each other (just like the sine
and cosine functions).

What is the significance of the


1 1 1 1 f1
1 1 1 1 f
c 2
1 1 0 0 f3

0 0 1 1 f4

f1 f 2 f3 f 4 The first coefficient


f f f3 f 4 c(1,1) is related to
c 1 2 the average value
f1 f2 of f, while the
remaining terms
f3 f4 correspond to high
frequency terms.
The form of the transformation matrix H
which we have used till now is in the un-
normalized form.

In the normalized form, each of the basis


functions has unit magnitude (or norm).

The normalized form of H4 is:


1 1 1 1
1 1 1 1
1
H4
4 2 2 0 0

0 0 2 2
1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1

2 2 2 2 0 0 0 0

1 0 0 0 0 2 2 2 2
H8
8 4 4 0 0 0 0 0 0

0 0 4 4 0 0 0 0
0 0 0 0 4 4 0 0

0 0 0 0 0 0 4 4

How do we implement the Haar wavelet


transform in two dimensions?

c HfH T
Here f is a N x N matrix (where N is a power
of 2) and H is also a N x N matrix.
H acts on the columns of f, while HT acts
on the rows of Hf.

How do we do the inverse


transformation?
Since H is an orthogonal matrix , HHT = I.
Therefore H-1 = HT
Therefore:

f H cH T
Let us check whether implementing
the Haar wavelet transform leads to
compression.
Input Image: Entropy = 7.002
bits/pixel
Size of image = 128 x 128
Apply the normalized Haar Wavelet
transform (of size 128 x 128).
The transformed figure looks like this!

Ghosts of the cameraman!


Entropy of transformed image = 5.7003
bits / pixel
Therefore, we can compress more!
Yes! We can.

Compare with the original image.

You might also like