HWTCoding

C ODING THE H AAR WAVELET T RANSFORM
Catherine Bénéteau, Caroline Haddad,

David Ruch and Patrick J. Van Fleet
Department of Mathematics & Statistics

University of South Florida
Tampa, FL USA
PREP - Wavelet Workshop, 2009
T UESDAY, 2 J UNE , 2009 (L ECTURE 4) C ODING THE HWT PREP 2009 1/9
T ODAY ’ S S CHEDULE
9:00-9:15 Welcome, Introductions

9:15-10:15 Lecture One: Why Wavelets?
10:15-10:30 Coffee Break (SCA 202)
10:30-11:45 Lecture Two: Digital Images
12:00-1:00 Lunch
1:30-2:45 Lecture Three: The Discrete Haar Wavelet Transform
3:00-4:15 ⇒Lecture Four: Coding the HWT, Edge Detection
O UTLINE
C ODING THE H AAR T RANSFORM

Implementing WN
Implementing WNT
Coding the 2D Transform
I N THE C LASSROOM
Teaching Ideas
Computer Usage
C ODING THE H AAR T RANSFORM I MPLEMENTING WN

I MPLEMENTING WN
I The natural inclination when coding the HWT is to simply write a

loop and compute the lowpass portion and the highpass portion in
the same loop.

I MPLEMENTING WN

the same loop.
I This bogs down in Mathematica and is also difficult to generalize
when we consider longer filters.

I MPLEMENTING WN

the same loop.
I This bogs down in Mathematica and is also difficult to generalize
when we consider longer filters.
I If we look at the lowpass portion of the transform, Hv, we can see
a better way to code things.

I MPLEMENTING WN
Consider H v when v ∈ R8 . We have
v1
 
 v2 
 1 1
    
0 0 0 0 0 0
2 2
 v3  v1 + v2
 0 0 1 1 0 0 0 0  
 
v4  1  v3 + v4 
Hv=
 2 2  · =  
0 0 0 0 12 12 0 0   v5  2  v5 + v6


0 0 0 0 0 0 21 12 
 v6 
 v7 + v8
 v7 
v8

I MPLEMENTING WN
If we rewrite this, we have

   
v1 + v2 v1 v2
1

1 v3 + v4   v3 v4 
· 2
Hv=  =
1 = Vh
2 v5 + v6
   v5 v6  2
v7 + v8 v7 v8

I MPLEMENTING WN
In a similar way we see that
v1
 
 v2 
− 12 12
    
0 0 0 0 0 0  v3  −v1 + v2
 0 0 −1 1
 
0 0 0 0  v4  1  −v3 + v4 
Gv =  2 2 · =  
 0 0 0 0 − 21 12 0 0   v5  2  −v5 + v6


0 0 0 0 0 0 − 21 12 
 v6 
 −v7 + v8
 v7 
v8

I MPLEMENTING WN
Rewriting gives
   
−v1 + v2 v1 v2 1
1 −v3 + v 4
  v3
 · −2 = Vg
v4 
Gv =  =
1
2  −v5 + v6   v5 v6  2
−v7 + v8 v7 v8
So all we need to do to compute Wn v is to create V , multiply it with h

and g, and join to the two blocks together!

I MPLEMENTING WN
Here is some Matlab code to do it:
function y=MyHWT1D1(v)
V=reshape(v,2,length(v)/2)’;
lp=V*[1/2,1/2];
hp=V*[-1/2,1/2];
y=[lp;hp];

I MPLEMENTING WN
And some Mathematica code as well:
MyHWT1D1[v_]:=Module[{V,lp,hp,y},
V=Partition[v,2];
lp=V.{1/2,1/2};
hp=V.{-1/2,1/2};
y=Join[lp,hp];
Return[y];
];
C ODING THE H AAR T RANSFORM I MPLEMENTING WNT

I MPLEMENTING WNT
I Writing the code for the inverse transform is a bit trickier.

I MPLEMENTING WNT

I Now the computation is
h i
WNT v = H T GT v


I MPLEMENTING WNT

I Now the computation is
h i
WNT v = H T GT v

I Let’s again look at a vector v ∈ R8 and consider the product W8T v:

I MPLEMENTING WNT
1 0 0 0 −1 0 0 0 v1
   

 1 0 0 0 1 0 0 0  
  v2 


 0 1 0 0 0 −1 0 0  
  v3 

T
 0 1 0 0 0 1 0 0  
· v4 
W8 v =  

 0 0 1 0 0 0 −1 0  
  v5 


 0 0 1 0 0 0 1 0  
  v6 

 0 0 0 1 0 0 0 −1   v7 
0 0 0 1 0 0 0 1 v8

I MPLEMENTING WNT
or
v1 − v5
 

 v1 + v5 


 v2 − v6 

T
 v2 + v6 
W8 v =  

 v3 − v7 


 v3 + v7 

 v4 − v8 
v4 + v8

I MPLEMENTING WNT
I The matrix V takes a bit different shape this time.

I MPLEMENTING WNT

I Now V is  
v1 v5
 v2 v6 
V =
 v3

v7 
v4 v8

I MPLEMENTING WNT

I Now V is  
v1 v5
 v2 v6 
V =
 v3

v7 
v4 v8
I We need to dot V with both h = [1, 1]T and g = [1, −1]T but then
intertwine the results.

I MPLEMENTING WNT
Given v = (v1 , . . . , v8 )T , how can we use Matlab/Mathematica

commands to form  
v1 v5
 v2 v6 
V = v3 v7 

v4 v8

I MPLEMENTING WNT
Answer (Matlab):
V = reshape(v,length(v)/2,2);

I MPLEMENTING WNT
Answer (Mathematica):
V = Transpose[Partition[v,Length[v]/2]];

I MPLEMENTING WNT
We can compute
   
v1 + v5 v1 − v5
 v2 + v6   v2 − v6 
Vh =  
 v3 + v7  and Vg =  
 v3 − v7 
v4 + v8 v4 − v8
How do we form W8T v from this?

I MPLEMENTING WNT
Answer (Matlab):
reshape([V*g V*h]’,1,length(v))’;

I MPLEMENTING WNT
Answer (Mathematica):
Flatten[Transpose[{ V g, V h } ]];
or one I just learned:
Riffle[ Vg, Vh ];

I MPLEMENTING WNT
Let’s have a look at the Mathematica notebook
HaarTransform1D.nb
to see how to code the one–dimensional HWT and its inverse.
C ODING THE H AAR T RANSFORM C ODING THE 2D T RANSFORM
2D H AAR T RANSFORM
C ODING THE 2D T RANSFORM
I Coding the 2D Haar transform is easy - we already have a routine

that will apply the HWT to each column of A,
2D H AAR T RANSFORM

I We can easily write a routine to compute C = WN A. Let’s call this
routine W.
2D H AAR T RANSFORM

routine W.
I Our goal is to compute B = WN AWNT = CWNT .
2D H AAR T RANSFORM

routine W.
I Our goal is to compute B = WN AWNT = CWNT .
I It turns out that writing code for CWNT is a bit tedious, but if we use
some linear algebra . . .
2D H AAR T RANSFORM
I If we transpose both sides of B = CWNT , we have
B T = WN C T
2D H AAR T RANSFORM
B T = WN C T
I So we can simply apply W to C T and transpose the result.
2D H AAR T RANSFORM
B T = WN C T
I So we can simply apply W to C T and transpose the result.

I One student wasn’t so sure about this . . .
2D H AAR T RANSFORM
I N THE C LASSROOM T EACHING I DEAS
I N THE C LASSROOM
T EACHING I DEAS
I The students really enjoy the material in this chapter. It is quite

straightforward and ties together everything new we’ve done to
date.
I N THE C LASSROOM T EACHING I DEAS
I N THE C LASSROOM
T EACHING I DEAS
I The students really enjoy the material in this chapter. It is quite

straightforward and ties together everything new we’ve done to
date.
I We have them look at the entropy of particular vectors when
processed by the Haar transform. This gives them some idea of
the potential for wavelet-based compression.
I N THE C LASSROOM C OMPUTER U SAGE
I N THE C LASSROOM
C OMPUTER U SAGE
I As you might imagine, we do lots of coding in this chapter.
I N THE C LASSROOM
C OMPUTER U SAGE

I We let the students work in pairs and they write code for the Haar
transform and its inverse (1D and 2D) as well as iterated versions.
I N THE C LASSROOM
C OMPUTER U SAGE

I We let the students work in pairs and they write code for the Haar
transform and its inverse (1D and 2D) as well as iterated versions.
I They can get pretty frustrated with Mathematica at this point - it is
good to show them some simple debugging techniques.
I N THE C LASSROOM
C OMPUTER U SAGE
I We let them use their own images/audio files (sometimes

dangerous).
I N THE C LASSROOM
C OMPUTER U SAGE
I We let them use their own images/audio files (sometimes

dangerous).
I To test their iterated 1D inverse, they must download an audio clip
from the course web site that has been transformed p times,
guess at p, and then apply their inverse to guess the movie clip.
9:00-9:15 Welcome, Introductions

9:15-10:15 Lecture One: Why Wavelets?
10:30-11:45 Lecture Two: Digital Images
12:00-1:00 Lunch
1:30-2:45 Lecture Three: The Discrete Haar Wavelet Transform
3:00-4:15 Lecture Four: Coding the HWT, Edge Detection
⇒Dinner and free time

HWTCoding

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

HWTCoding

Uploaded by

Copyright:

Available Formats

C ODING THE H AAR WAVELET T RANSFORM

Catherine Bénéteau, Caroline Haddad,

Department of Mathematics & Statistics

PREP - Wavelet Workshop, 2009

9:00-9:15 Welcome, Introductions

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

I The natural inclination when coding the HWT is to simply write a

C ODING THE H AAR T RANSFORM

I The natural inclination when coding the HWT is to simply write a

C ODING THE H AAR T RANSFORM

I The natural inclination when coding the HWT is to simply write a

C ODING THE H AAR T RANSFORM

Consider H v when v ∈ R8 . We have

C ODING THE H AAR T RANSFORM

If we rewrite this, we have

C ODING THE H AAR T RANSFORM

In a similar way we see that

C ODING THE H AAR T RANSFORM

So all we need to do to compute Wn v is to create V , multiply it with h

C ODING THE H AAR T RANSFORM

Here is some Matlab code to do it:

C ODING THE H AAR T RANSFORM

And some Mathematica code as well:

C ODING THE H AAR T RANSFORM

I Writing the code for the inverse transform is a bit trickier.

C ODING THE H AAR T RANSFORM

I Writing the code for the inverse transform is a bit trickier.

C ODING THE H AAR T RANSFORM

I Writing the code for the inverse transform is a bit trickier.

I Let’s again look at a vector v ∈ R8 and consider the product W8T v:

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

I The matrix V takes a bit different shape this time.

C ODING THE H AAR T RANSFORM

I The matrix V takes a bit different shape this time.

C ODING THE H AAR T RANSFORM

I The matrix V takes a bit different shape this time.

C ODING THE H AAR T RANSFORM

Given v = (v1 , . . . , v8 )T , how can we use Matlab/Mathematica

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

How do we form W8T v from this?

C ODING THE H AAR T RANSFORM

C ODING THE H AAR T RANSFORM

or one I just learned:

C ODING THE H AAR T RANSFORM

Let’s have a look at the Mathematica notebook

to see how to code the one–dimensional HWT and its inverse.

I Coding the 2D Haar transform is easy - we already have a routine

I Coding the 2D Haar transform is easy - we already have a routine

I Coding the 2D Haar transform is easy - we already have a routine

I Coding the 2D Haar transform is easy - we already have a routine

I If we transpose both sides of B = CWNT , we have

I If we transpose both sides of B = CWNT , we have

I So we can simply apply W to C T and transpose the result.

I If we transpose both sides of B = CWNT , we have

I So we can simply apply W to C T and transpose the result.

I The students really enjoy the material in this chapter. It is quite

I The students really enjoy the material in this chapter. It is quite

I As you might imagine, we do lots of coding in this chapter.

I As you might imagine, we do lots of coding in this chapter.