Homework:
Performance of DCE compared to subsampling
Does the DCE really perform better than simple subsampling
in the sense of the Hausdorff distance ?
Please program the following experiment:
For all shapes in a given testset:
Create simplified versions having 103,98,93,88,..,3 vertices using
- DCE
- Subsampling with equal runlength distance
Measure the distance to the original object using the
polygonal Hausdorff distance HP, which is defined
in the following way:
Let S be the original shape, A the simplified (abstracted)
version. Then
HP=max(min(d(S,ai)) for all vertices ai of A,
d(S,ai) being the set of euclidean distance between ai and all segments
of S (distance point-segment = shortest euclidean distance).
Draw 3 diagrams showing on the x axis the number of
vertices in A, on the y axis the distance value.
Draw curves for
- the average,
- min and
- max distance over all shapes
for both distance measures.
You will be given matlab programs for:
DCE, subsampling, distance point-segment, tools (loading/saving/plotting),
and a set of shapes. The set of shapes is the 1400 shapes used for the
MPEG-7 test. It is more or less the industrial standard set for tests regarding
shape.
You get it all HERE (about 4MB)