Wave Letghchgch

26
The Discrete Wavelet Transform for Image Compression Speaker: Jing-De Huang Advisor: Jian-Jiun Ding Graduate Institute of Communication Engineering National Taiwan University, Taipei, Taiwan, ROC

description

hh

Transcript of Wave Letghchgch

Page 1: Wave Letghchgch

The Discrete Wavelet Transformfor Image Compression

Speaker: Jing-De HuangAdvisor: Jian-Jiun Ding

Graduate Institute of Communication EngineeringNational Taiwan University, Taipei, Taiwan, ROC

Page 2: Wave Letghchgch

2

Outline

• Subband Coding

• Multiresolution Analysis

• Discrete Wavelet Transform

• The Fast Wavelet Transform

• Wavelet Transforms in Two Dimension

• Image Compression

• Simulation Result

Page 3: Wave Letghchgch

3

1. Subband Coding

h0(n)

h1(n)

2

2

2

2

g0(n)

g1(n)

+Analysis Synthesis

1( )y n

0 ( )y n

( )x n ˆ( )x n

1( )H 1( )H

/ 2

Low band High band

0

Page 4: Wave Letghchgch

4

1. Subband Coding

• Cross-modulated

10 0 1 12

10 0 1 12

ˆ ( ) ( ) ( ) ( ) ( ) ( )

( ) ( ) ( ) ( ) ( )

X z H z G z H z G z X z

H z G z H z G z X z

ˆ ( ) ( )X z X z

0 0 1 1

0 0 1 1

( ) ( ) ( ) ( ) 0

( ) ( ) ( ) ( ) 2

H z G z H z G z

H z G z H z G z

For finite impulse response (FIR) filters and ignoring the delay

0 1

11 0

( ) ( 1) ( )

( ) ( 1) ( )

n

n

g n h n

g n h n

For error-free reconstruction

FIR synthesis filters are cross-modulated copies of the analysis filters with one (and only one) being sign reversed.

Z-transform :

Page 5: Wave Letghchgch

5

1. Subband Coding

• Biorthogonal

(2 ), ( ) ( ) ( ), , {0,1}i jh n k g k i j n i j

0 0 0 1

1 1 1 0

( ), (2 ) ( ), ( ), (2 ) 0

( ), (2 ) ( ), ( ), (2 ) 0

g k h n k n g k h n k

g k h n k n g k h n k

The analysis and synthesis filter impulse responses of all two-band,

real-coefficient, perfect reconstruction filter banks are subject to the biorthogonality constraint

Page 6: Wave Letghchgch

6

1. Subband Coding

• Orthonormal – – one solution of biorthogonal– used in the fast wavelet transform– the relationship of the four filter is :

( ), ( 2 ) ( ) ( ), , {0,1}i jg n g n m i j m i j

1 0

1 0

1 2 1 2

( ) ( 1) (2 1 ) , 2K denotes the number of coefficients

( ) is related to ( )

( ) (2 1 ), {0,1}

( ), ( ) is time-reversed versions of ( ), ( ), respectively

n

i i

g n g K n

g n g n

h n g K n i

h n h n g n g n

Page 7: Wave Letghchgch

7

2. Multiresolution Analysis

( ) ( )k kk

f x x

• Expansion of a signal f (x) :: real-valued expansion coefficients

( ) : real-valued expansion functionsk

k x

If the expansion is unique, the are called basis functions. ( )k x

The function space of the expansion set :( )k x ( )kk

V span x

*( ), ( ) ( ) ( ) ( ): the dual function of ( )k k k k kx f x x f x dx x x

If is an orthonormal basis for V , then ( )k x ( ) ( )k kx x

If are not orthonormal but are an orthogonal basis for V , then the basis funcitons and their duals are called biorthogonal.

( )k x

0 ,Biorthogonal: ( ), ( )

1 ,j k jk

j kx x

j k

Page 8: Wave Letghchgch

8

2. Multiresolution Analysis• Scaling function

/ 2 2, ( ) 2 (2 ), for and ( )j j

j k x x k k x L Z R

The subspace spanned over k for any j : , ( )j j kk

V span x

The scaling functions of any subspace can be built from double-resolution copies of themselves. That is,

where the coefficients are called scaling function coefficients.

( ) ( ) 2 (2 )n

x h n x n

Page 9: Wave Letghchgch

9

2. Multiresolution Analysis

• Requirements of scaling function:

1. The scaling function is orthogonal to its integer translates.

2. The subspaces spanned by the scaling function at low scales are nested within those spanned at higher scales. That is

3. The only function that is common to all is .That is

4. Any function can be represented with arbitrary precision. That is,

1 0 1 2V V V V V V

jV ( ) 0f x

0V

2V L R

Page 10: Wave Letghchgch

10

2. Multiresolution Analysis

• Wavelet function

/ 2, ( ) 2 (2 )j j

j k x x k

spans the difference between any two adjacent scaling subspaces and

jV 1jV

for all that spans the space k Z jW

where , ( )j j kk

W span x

The wavelet function can be expressed as a weighted sum of shifted, double-resolution scaling functions. That is,

where the are called the wavelet function coefficients.

( ) ( ) 2 (2 )n

x h n x n ( )h n

It can be shown that ( ) ( 1) (1 )nh n h n

Page 11: Wave Letghchgch

11

2. Multiresolution Analysis

2 1 1 0 0 1V V W V W W

0V0W1W

1 0 0V V W

Figure 2 The relationship between scaling and wavelet function spaces.

The scaling and wavelet function subspaces in Fig. 2 are related by

We can express the space of all measurable, square-integrable function as

or

1j j jV V W

20 0 1 2L V W W W R

22 1 0 1 2L W W W W W R

Page 12: Wave Letghchgch

12

3 Discrete Wavelet Transform

• Wavelet series expansion

0 0

0

, ,( ) ( ) ( ) ( ) ( )j j k j j kk j j k

f x c k x d k x

where j0 is an arbitrary starting scale

0 0 0, ,( ) ( ), ( ) ( ) ( )j j k j kc k f x x f x x dx

, ,( ) ( ), ( ) ( ) ( )j j k j kd k f x x f x x dx

called the approximation or scaling coefficients

called the detail or wavelet coefficients

Page 13: Wave Letghchgch

13

3 Discrete Wavelet Transform

• Discrete Wavelet Transform

where j0 is an arbitrary starting scale

called the approximation or scaling coefficients

called the detail or wavelet coefficients

0

0

0 , ,

1 1( ) ( , ) ( ) ( , ) ( )j k j k

k j j k

f x W j k x W j k xM M

the function f(x) is a sequence of numbers

0

1

0 ,0

1( , ) ( ) ( )

M

j kx

W j k f x xM

1

,0

1( , ) ( ) ( )

M

j kx

W j k f x xM

Page 14: Wave Letghchgch

14

4 The Fast Wavelet Transform

• Fast Wavelet Transform (FWT) – computationally efficient implementation of the DWT– the relationship between the coefficients of the DWT at adjacent

scales – also called Mallat's herringbone algorithm – resembles the twoband subband coding scheme

Page 15: Wave Letghchgch

15

4 The Fast Wavelet Transform

( ) ( ) 2 (2 )n

x h n x n Scaling x by 2j, translating it by k, and letting m = 2k + n

1(2 ) ( ) 2 2(2 ) ( 2 ) 2 2j j j

n m

x k h n x k n h m k x m

1(2 ) ( 2 ) 2 (2 )j j

m

x k h m k x m Similarity,

Consider the DWT. Assume and ( ) ( )x x ( ) ( )x x

( ) ( ) 2 (2 )n

x h n x n 0

0

1

0 ,0

1( , ) ( ) ( )

1( 2 ) 2 (2 )

M

j kx

j

m

W j k f x xM

h m k x mM

Page 16: Wave Letghchgch

16

4 The Fast Wavelet Transform / 2

, ( ) 2 (2 )j jj k x x k

1

,0

/ 2

/ 2 1

( 1) / 2 1

1( , ) ( ) ( )

1( )2 (2 )

1( )2 ( 2 ) 2 (2 )

1( 2 ) ( )2 (2 )

( 2 ) ( 1, )

M

j kx

j j

x

j j

x m

j j

m x

m

W j k f x xM

f x x kM

f x h m k x mM

h m k f x x mM

h m k W j m

Similarity, ( , ) ( 2 ) ( 1, )

m

W j k h m k W j m

Page 17: Wave Letghchgch

17

2 , 0

2 , 0

( , ) ( ) ( 1, )

( , ) ( ) ( 1, )

n k k

n k k

W j k h n W j n

W j k h n W j n

Figure 3 An FWT analysis filter bank.

4 The Fast Wavelet Transform

Page 18: Wave Letghchgch

18

4 The Fast Wavelet Transform

Figure 4 An FWT-1 synthesis filter bank.

By subband coding theorem, perfect reconstrucion for two-band orthonormal filters requires for i = {0, 1}. That is, the synthesis and analysis filters must be time-reversed versions of one another. Since the FWT analysis filter are and , the required FWT -1 synthesis filtersare and .

( ) ( )i ig n h n

0 ( ) ( )h n h n 1( ) ( )h n h n

0 0( ) ( ) ( )g n h n h n 1 1( ) ( ) ( )g n h n h n

Page 19: Wave Letghchgch

19

Wavelet Transform vs. Fourier Transform

• Fourier transform– Basis function cover the entire signal range,

varying in frequency only

• Wavelet transform– Basis functions vary in frequency (called “scale”)

as well as spatial extend• High frequency basis covers a smaller area• Low frequency basis covers a larger area

Page 20: Wave Letghchgch

20

Wavelet Transform vs. Fourier Transform

Time-frequency distribution for (a) sampled data, (b) FFT, and (c) FWT basis

Page 21: Wave Letghchgch

21

5 Wavelet Transforms in Two Dimension

Figure 5 The two-dimensional FWT the analysis filter.

Page 22: Wave Letghchgch

22

5 Wavelet Transforms in Two Dimension

( 1, , )W j m n

( , , )W j m n ( , , )HW j m n

( , , )VW j m n ( , , )DW j m n

Figure 6 Two-scale of two-dimensional decomposition

( 1, , )W j m n

( , , )W j m n ( , , )HW j m n

( , , )VW j m n ( , , )DW j m n

two-dimensional decomposition

Page 23: Wave Letghchgch

23

5 Wavelet Transforms in Two Dimension

Page 24: Wave Letghchgch

24

5 Wavelet Transforms in Two Dimension

Figure 7 The two-dimensional FWT the synthesis filter bank.

Page 25: Wave Letghchgch

25

6 Image Compression

Wavelet coding

QuantizationEntropy coding

image bitstream

• Quantization– uniform scalar quantization – separate quantization step-sizes for each subband

• Entropy coding– Huffman coding– Arithmetic coding

( , )( , ) sign( ( , ))

j

j jj

W m nq m n y m n

Page 26: Wave Letghchgch

26

7 Simulation Result

I1: Original image with width W and height H

C: Encoded jpeg stream from I1 I2: Decoded image from C

CR (Compression Ratio) = sizeof(I1) / sizeof(C)

RMS (Root mean square error) =

I1 C I2encoder decoder

2

1 21 1

( , ) ( , ) /( )H W

y x

I x y I x y H W