age.
lage.
Figure 10: MAP-refined filtered SAR image.
3.2 Image Processing Filters
In addition to the SAR-specific filters in EV-SAR, the
standard EarthView commands include more general
filtering techniques. Both median filtering and lower-
upper-middle (LUM) filtering can be effective in re-
moving speckle and other noise. The LUM filter is
a rank ordered filter and is capable of simultaneously
smoothing and sharpening different features in an im-
age. As well, EarthView provides spatial and spectral
convolution filtering utilizing finite impulse response
(FIR) or boxcar techniques and FFTs. These provide
the basics for removing or enhancing specific image
content.
3.3 Histogram Modification
Image enhancement can be achieved by modifying
image data such that it more closely matches a desired
distribution. Often it is desirable to stretch a particular
region of intensity values to provide more contrast.
Also, the expected distribution of an image may be
known. EarthView provides the capability to modify
image data to more closely match a uniform, cube root,
logarithmic, exponential or Rayleigh distribution.
3.4 SAR Image Classification
EarthView provides standard multispectral classifi-
cation methods including box, minimum-distance and
maximum-likelihood. A training function is provided
allowing the selection of multiple polygonal regions for
multiple class types. Texture based classification can
also be performed using FarthView. This classification
is based on the texture unit described by He and Wang
[3].
4. SAR IMAGE DISPLAY AND MEASUREMENT
SAR data sets are typically very large. One scene of
100kmx 100km at 25m resolution is 8000x8000 pixels,
16 bits per pixel, for 122 MB of data. The large
data sets combined with the high dynamic range of
the imagery requires special processing. Since all
8000x8000 pixels cannot be displayed on screen at
once, EV-SAR displays images using automatic pixel
averaging to low pass filter and resample the image
data to conform to the display screen pixel dimensions.
Dynamic range of the images are rescaled for the gray
scale range of the display. To access the full resolution
of the data, interactive image zooming and panning is
available. To enhance interpretation of the data values
pixel value readoff is possible as is latitude/longitude
tagging to find a pixel at a specific geodetic coordinate.
4.1 Complex Image Display
One special type of SAR imagery is complex imagery
where each pixel value is a complex number with a
magnitude and phase. The phase component contains
fine range information. When two SAR images are
multiplied pixel by pixel, an interferogram is made
containing a fringe pattern which may be phase un-
wrapped to derive terrain elevation information. It is
thus necessary to display the complex image magni-
tude and phase either separately or combined at once.
To optimize the display of complex imagery, EV-SAR
maps the complex image magnitude into image bright-
ness and the phase component into a colour wheel of
red merging to green, then blue, and finally back to red
again completing the circle. Optionally, a phase-only
image can be displayed. An example of an interfer-
ometric SAR image is given in Figure 11 where the
magnitude and phase are shown in two separate im-
ages.
4.2 Point Target Response Analysis
Determining the quality of a SAR image is important
for correctly interpreting and analyzing the data. This
is particularly important when the imagery has been
post processed using spacial filters and resampling.
The most accepted method for measuring SAR im-
age quality is the analysis of the point target response
of a strong point target, such as a calibration reflec-
tor, against a radiometrically dark background. Point
target measurements in EV-SAR are made by interac-
tively windowing a point target in a displayed image,
and then running an analysis on the target. The analy-
sis function interpolates the point target response, and
produces a three-dimensional plot of the point target
303
