|
Image Fusion
Since both images are registered, the boundaries between
structures in the scene should be identical. This implies that the
edges in the segmentations should be similar, even though the mean
value within each segment will differ.
For image fusion we perform a joint two-dimensional segmentation
of both images. The total likelihood for independent processes is
simply the sum of the individual contributions. The resulting joint
segmentation should show improved boundary definition compared with
the individual results.
Image fusion could also involve joint classification which should
yield better performance.
The advantage of supervised classification is that we can assess
this improvement by observing the classification performance
compared to the ground truth.
|
