Multi-Resolution Analysis

(MRA)
 FFT Vs Wavelet
• FFT, basis functions: sinusoids
• Wavelet transforms: small waves, called
wavelet
• FFT can only offer frequency information
• Wavelet: frequency + temporal information
• Fourier analysis doesn’t work well on
discontinuous, “bursty” data
– music, video, power, earthquakes,…
 Fourier versus Wavelets
• Fourier
– Loses time (location) coordinate completely
– Analyses the whole signal
– Short pieces lose “frequency” meaning

• Wavelets
– Localized time-frequency analysis
– Short signal pieces also have significance
– Scale = Frequency band
 Wavelet Definition
“The wavelet transform is a tool that cuts up
data, functions or operators into different
frequency components, and then studies
each component with a resolution matched
to its scale”
Dr. Ingrid Daubechies, Lucent, Princeton U
 Fourier transform
Fourier transform:
 Continuous Wavelet transform
for each Scale


for each Position
Coefficient (S,P) = Signal x Wavelet (S,P)
all time
end
end

Coefficient Scale
 Wavelet Transform
• Scale and shift original waveform
• Compare to a wavelet
• Assign a coefficient of similarity
 Scaling-- value of “stretch”
• Scaling a wavelet simply means
stretching (or compressing) it.
f(t) = sin(t) f(t) = sin(3t)
f(t) = sin(2t)
scale factor 3
scale factor1 scale factor 2
 More on scaling
• It lets you either narrow down the frequency band
of interest, or determine the frequency content in a
narrower time interval
• Scaling = frequency band
• Good for non-stationary data
• Low scalea Compressed wavelet
Rapidly changing detailsHigh frequency .
• High scale a Stretched wavelet  Slowly
changing, coarse features  Low frequency
Scale is (sort of) like frequency

Small scale
-Rapidly changing details,
-Like high frequency

Large scale
-Slowly changing
details
-Like low frequency
Scale is (sort of) like frequency

The scale factor works exactly the same with

wavelets. The smaller the scale factor, the
more "compressed" the wavelet.
 Shifting
Shifting a wavelet simply means delaying (or hastening) its
onset. Mathematically, delaying a function  f(t)   by k is
represented by f(t-k)
Shifting

C = 0.0004

C = 0.0034
 Five Easy Steps to a Continuous
Wavelet Transform
1. Take a wavelet and compare it to a section at the start of
the original signal.
2. Calculate
1. a correlation coefficient c

 
       
2.S
                                                                                                                                                                                            
 Five Easy Steps to a Continuous
Wavelet Transform
3. Shift the wavelet to the right and repeat steps 1 and 2 until
you've covered the whole signal.
4. Scale (stretch) the wavelet and repeat steps 1 through 3.
5. Repeat steps 1 through 4 for all scales.
Coefficient Plots
 Discrete Wavelet Transform
• “Subset” of scale and position based on
power of two
– rather than every “possible” set of scale and
position in continuous wavelet transform
• Behaves like a filter bank: signal in,
coefficients out
• Down-sampling necessary (twice as much
data as original signal)
 Discrete Wavelet transform
signal

lowpass highpass

filters

Approximation Details
(a) (d)
 Results of wavelet transform:
approximation and details
• Low frequency:
– approximation (a)

• High frequency
– Details (d)

• “Decomposition”
can be performed
iteratively
Levels of decomposition

• Successively decompose
the approximation

• Level 5 decomposition =
a5 + d5 + d4 + d3 + d2 +
d1

• No limit to the number of

decompositions performed
 Wavelet synthesis

•Re-creates signal from coefficients

•Up-sampling required
 Multi-level Wavelet Analysis
Multi-level wavelet
decomposition tree Reassembling original signal
Non-stationary Property of
Natural Image
Pyramidal Image Structure
 Image Pyramids
• Original image, the base of the pyramid, in the
level J =log N, Normally truncated to P+1 levels.
2

• Approximation pyramids, predication residual

pyramids
• Steps: .1. Compute a reduced-resolution
approximation (from j to j-1 level) by
downsampling; 2. Upsample the output of step1,
get predication image; 3. Difference between the
predication of step 2 and the input of step1.
Subband Coding
 Subband Coding
• Filters h1(n) and h2(n) are half-band digital
filters, their transfer characteristics H 0-low
pass filter, output is an approximation of
x(n) and H1-high pass filter, output is the
high frequency or detail part of x(n)
• Criteria: h0(n), h1(n), g0(n), g1(n) are
selected to reconstruct the input perfectly.
 Z-transform
• Z- transform a generalization of the discrete
Fourier transform
• The Z-transform is also the discrete time
version of Laplace transform
• Given a sequence{x(n)}, its z-transform is
• X(z) = 

x(n) z n
 Subband Coding
^
1 1
X ( z) 2 0 0
 [ H ( z )G ( z )  H 1 ( z )G1 ( z )] X ( z ) 
2
[ H 0 ( z )G0 ( z )  H1 ( z )G1 ( z )] X ( z )

H 0 (  z )G0 ( z )  H 1 (  z )G1 ( z )  0
H 0 ( z )G0 ( z )  H1 ( z )G1( z )  2
2-D 4-band filter bank
Approximation

Vertical detail

Horizontal detail

Diagonal details
Subband Example
 Haar Transform
Haar transform, separable and symmetric
T = HFH, where F is an NN image matrix
H is NN transformation matrix, H contains
the Haar basis functions, hk(z)
H0(t) = 1 for 0  t < 1  1
1if 0  t  2
H1 (t )  
1
  1if
 2
 n (n  0.5)
 2 p for p  t 
2 2p
 (n  0.5) (n  1)
H 2 p  n (t )   2 p for p
t 
 2 2p
 0otherwise

whereP  1,2,3,...andn  0,1,...,2 p  1
 Haar Transform
 1
1if 0  t  2
H 1 (t )  
1
  1if
 2
 n (n  0.5)
 2 p
for  t 
2p 2p
 (n  0.5) (n  1)
H 2 p  n (t )   2 for
p
p
t  p
 2 2
 0otherwise

whereP  1,2,3,...andn  0,1,...,2 p  1
 Series Expansion
• In MRA, scaling function to create a series of
approximations of a function or image, wavelet to
encode the difference in information between
different approximations
• A signal or function f(x) can be analyzed as a
linear combination of expansion functions
f ( x)    k k
k

 k  exp ansioncoefficients
 k ( x)  basisfunction
 Scaling Function
Set{j,k(x)} where,
 j ,k ( x)  2 j / 2  (2 j x  k ) forallj , k  Z

K determines the position of j,k(x) along the

x-axis, j -- j,k(x) width, and 2j/2—height or
amplitude
The shape of j,k(x) change with j, (x) is
called scaling function
Haar scaling function
 Fundamental Requirements of
MRA
• The scaling function is orthogonal to its integer
translate
• The subspaces spanned by the scaling function at
low scales are nested within those spanned at
higher scales
• The only function that is common to all Vj is f(x)
=0
• Any function can be represented with arbitrary
precision
 Refinement Equation
 ( x)  0,0 ( x)
 ( x)   h (n) 2 (2 x  n)
n

h(x) coefficient –scaling function coefficient

h(x) – scaling vector
The expansion functions of any subspace can
built from the next higher resolution space
Wavelet Functions
Wavelet Functions
Wavelet Function
2-D Wavelet Transform
Wavelet Packets
 2-D
Wavelets
 Applications of wavelets
• Pattern recognition
– Biotech: to distinguish the normal from the pathological
membranes
– Biometrics: facial/corneal/fingerprint recognition
• Feature extraction
– Metallurgy: characterization of rough surfaces
• Trend detection:
– Finance: exploring variation of stock prices
• Perfect reconstruction
– Communications: wireless channel signals
• Video compression – JPEG 2000
 Useful Link
• Matlab wavelet tool using guide
• http://www.wavelet.org
• http://www.multires.caltech.edu/teaching/
• http://www-dsp.rice.edu/software/RWT/
• www.multires.caltech.edu/teaching/courses/
waveletcourse/sig95.course.pdf
• http://www.amara.com/current/wavelet.html