Consider an image of dimension 100 x 100. Compare the
memory requirements to represent the data using
a structured point dataset
an unstructured point dataset
When designing an interpolation kernel, the
various issues that one must consider
are:
the width of the kernel (i.e., the number of sampled points
to be covered by the kernel);
any additional free parameters to be associated with the
kernel function;
the kernel function itself.
In the last issue above, it is important that the kernel function
is so designed
that the sum of all the function values evaluated at the sampled
points is 1, as these function values would be used as weights
in the interpolation process. Ideally, the kernel
function should be continuous and smooth also.
The cubic interpolation kernel that is defined below:
f(x) =
(a + 2) |x|3 - (a
+ 3) |x|2 + 1
if 0 <= |x| <= 1
a |x|3 - 5a |x|2
+ 8a |x| - 4a
if 1 <= |x| <= 2
0
if 2 <= |x|
has a free parameter, a.
Verify that the kernel gives
w0 +
w1 +
w2 +
w3 = 1, where
w0,
w1,
w2, and
w3
are the weights at the sampled points
x0=-1.5,
x1=-0.5,
x2=0.5, and
x3=1.5.
Is the total weight equal to 1 regardless of
the free paramerer a?
The B-spline interpolation kernel is given below:
f(x) =
1/2 |x|3 - |x|2
+ 2/3
if 0 <= |x| <= 1
-1/6 |x|3 + |x|2
- 2|x| + 4/3
if 1 <= |x| <= 2
0
if 2 <= |x|
Show that the kernel function is C1.
Show that the sum of the weights at
x0=-1.5,
x1=-0.5,
x2=0.5, and
x3=1.5
is 1.
Show that the linearly interpolated attribute value,
f(Q), of any point Q inside the
triangle formed by 3 sampled points
P0,
P1, and
P2
is
f(Q) = w0f(P0) +
w1f(P1) +
w2f(P2)
where
w0 = (1-r-s)
w1 = r
w2 = s
The two scalars r and s are defined as follows: r= ||P0Q1|| / ||P0P1 || , s= ||P0Q2|| / ||P0P2||, where QQ1is parallel to P0P2 and QQ2is parallel to P0P1, as shown in
the diagram
below: