Return to the homepage
Search the complete Wavelet Digest database
Help about the Wavelet Digest mailing list
About the Wavelet Digest
The Wavelet Digest
> Volume 6, Issue 4
Answer: Smoothest Scaling Function (WD 6.3 #21)

Previous :: Next

Author 
Message 
unser@helix.nih.gov (Michael Unser) Guest

Posted: Thu Mar 06, 1997 4:02 pm Subject: Answer: Smoothest Scaling Function (WD 6.3 #21)




#18 Answer: Smoothest Scaling Function (WD 6.3 #21)
>What is the smoothest scaling function, w(x), solving a
>'wavelet equation' of type:
>(1) w(x) = c(0)w(2x) + c(1)w(2x1) + c(2)w(2x2) +...+ c(m)w(2xm),
>where the coefficients c(i) satisfy the 'usual conditions':
>(2) sum c(i) over odd i = 1,
>(3) sum c(i) over even i = 1,
>and where w is normalized to have, say, integral = 1.
First, the smoothness of a scaling function cannot be greater than p
(the order of the function = its ability to reproduce polynomials of
degree p1 = corresponds to the factor (1+z)^p in the refinement
filter C(z)=(1+z)^p*Q(z)).
Second, it is easy to see that the Bspline of order p (or degree p1)
is the shortest scaling function of order p, because Q(z)=1.
Third, Bsplines of order p have a Sobolev smoothness index of p1/2
(cf. Strang's book). However, Bsplines are not the smoothest scaling
functions of order p! For instance, one can take Q(z)=(1+EPSILON+z),
which for EPSILON>0 sufficiently small (but nonzero !) can achieve
the maximum smoothness s_max=p. But this also shows that the Bsplines
are the smoothest scaling for a refinement filter of a given length:
the example above with EPSILON=0 is a Bspline of order p+1 which has
a smoothness s_max=p+1/2  this is 1/2 better than anything else of
the same length !
Michael Unser
BEIP, Building 13/3N17
National Institutes of Health
Bethesda, MD 208925766, USA
Email : Unser@helix.nih.gov 





All times are GMT + 1 Hour

Page 1 of 1 
