1.
Se ~
PT ove
pos Ww ^ 9^ A Mi
Figure 2 - Feature Boundaries
The output. from the outline algorithm consisted of the
feature boundary and any accompanying operator-identified
points of interest, all expressed as row and column pixel
values in the local picture coordinate system. All
additional picture information was discarded, reducing the
amount of feature information to a manageable amount. Also,
the background noise was greatly reduced. Obviously, this
method would not function well in image regions of low
contrast and smooth texture - it is feature based and
requires well-defined feature boundaries.
A general purpose transformation routine was developed to
match feature outlines from film to film. It was based on a
standard six parameter affine transformation, with
parameters for scaling on the x and y axes, translation in x
and y, rotation, and non-orthogonality of the axes. Points
in System 1 can be transformed to System 2 by:
cosá 0
cosf -sinf S, X, T
sind cosé
sinß cosf S, Yi Ty
Scale on the x axis
Scale on the y axis
where S
5,
ß = Rotation
6
T
T
= Non-orthogonality
= Translation in x
= Translation in y
Note that this transformation is usually implemented with
- 167 -