For
α>-1
α
-1
, B-splines can be extended from having order
nn, a positive integer, to a family
indexed by a fractional parameter. These generalized B-splines
are called fractional B-splines.
We will assume equally spaced knots with unit spacing. Remember
that B-splines are constructed from the (
n+1
n
1
) fold convolution of
β
0
x
β
0
x
; they are a basis for polynomial splines. We will use
the uncentered version of B-splines:
β
+
0
x=1ifx∈010.5ifx=0∨10otherwise
β
+
0
x
1
x
0
1
0.5
x
0
1
0
β
+
0
x=
β
+
,
1
0
*
β
+
,
2
0
*…*
β
+
,
n
0
*
β
+
,
n
+
1
0
x
β
+
0
x
β
+
,
1
0
β
+
,
2
0
…
β
+
,
n
0
β
+
,
n
+
1
0
x
A B-spline of order nn has the form
β
+
n
x=1n!∑k=0n+1n+1k-1kx-kn
β
+
n
x
1
n
k
0
n
1
n
1
k
1
k
x
k
n
(1)
and Fourier transform
β
^
n
ω=1-ⅇ-ⅈωⅈωn+1
β
^
n
ω
1
ω
ω
n
1
(2)
A fractional B-spline of order
αα has the form
β
+
α
x=1Γα+1∑k=0∞α+1k-1kx-kα
β
+
α
x
1
Γ
α
1
k
0
α
1
k
1
k
x
k
α
(3)
and Fourier transform
β
^
+
α
x=1-ⅇ-ⅈωⅈωα+1
β
^
+
α
x
1
ω
ω
α
1
(4)
One can see that the two definitions are very similar. However,
since the formula for computing binomial coeffients only allows
for integers, we must substitute generalized binomials.
These fractional B-splines interpolate conventional B-splines,
which are fractional B-splines of integer values. For a picture
of this, see a demonstration on Michael Unser's webpage.
Michael Unser and Theirry Blu developed fractional splines.