Speckle filters
Several image filtering methods for speckle removal have been tested.
The tested methods are
- Principal component analysis (PCA) for multitemporal images,
- Epos filtering (Hagg 1994)
- Iterated Conditional Modes method (ICM) based on Gibbsian random
field assumption
- Sigmafilter
- Simulated annealing method.
EPOS
EPOS-filter, Edge Preserving Optimized Speckle,
consists of four phases: estimation of the relative standard
deviation R from the image, calculation of the relative
standard deviation R(f) for all degrees f of freedom from
the chi-squared distribution, searching for the largest
homogeneous area around each pixel and calculating the new
grey value for the area by averaging the original values.
Below are two images, first unprocessed ERS-1 SAR image and
then EPOS-filtered one.
ICM
ICM method have been applied on one image at time with
a window size of 3 x 3 pixels and with about three iterations.
Regions approximately of the size of a typical stand can be recognized
after the iterations.
Sigmafilter
Sigmafilter is an edge-preserving speckle removing filter introduced
by J.S. Lee.
The method is simple but fairly effective and operates on one image at time.
Simulated annealing
Simulated annealing method has also been tested.
Results were fairly realistically segmented images,
but the size of the segments was too big to be of use in
operative forest management or removal of image speckle.
Contact information
- Finnish Forest Research Institute METLA
- Unioninkatu 40A, FIN-00170 Helsinki, Finland
- tel. +358-0-857 051, fax +358-0-625 308
- Internet: Erkki.Tomppo@metla.fi, Petri.Mikkela@metla.fi
[METLA] [National Forest Inventory]
[ERS1-SAR Project]
PM, August 30, 1995.