Next we investigate applying classification
after an adaptive filter is applied to the data, in the form of
segmentation. This offers the advantage of reducing speckle noise
by averaging over homogeneous segments. A variety of polarisation
measures can be used in segmentation.
 |
 |
| (c) The span measure
comprises the sum of the intensities in each polarisation.
When this is evaluated over a 3x3 window and then segmented
the correct classification probability rises to 85.1%. |
(d) Further
improvement is observed when the determinant is evaluated
over the 3x3 window and then segmented. The probability of
correct classification rises to 89.2%. |
|
|
|
