t
.dera&-
> pre-
rived
tions
;
ue to
| image
y»
1g
helpful in removing noise from an image, such as might be introduced by
electronic defects in the sensor. Figure 1 a shows an aircraft image with
coherent noise, followed by the final version of the image after removing
the noise (Fig. 1 b). Low frequency notch filtering can be useful removing
large scale shading which obscures more interesting local detail. This pro-
cedure is most frequently implemented in terms of a subtractive box filter,
in which a local average value is subtracted from each pixel (e.g. Soha
et al., 1977). Care must be exercised to avoid artifacts (Gillespie, 1976).
High frequency boost filtering can be used to sharpen edges. Where sensor
characteristics are known a priori, including noise level, Wiener filtering
can be employed to quantitatively restore image sharpness with minimum mean
square error (Helstrom,1967; Arp and Lorre,1976). Figure 2 demonstrates re-
storation filtering applied to a Mariner 10 view of Earth.
Figure 2: Example of restoration filtering applied to Mariner 10 view
of the Earth: (a) hefore, (b) after filtering.
3.2 Geometric Rectification
Imagery is mostly acquired with a geometry that needs correction for
efficient image analysis. The main purpose of rectification is to relate the
image to other data, be they in the form of maps, of other images or of
non-images such as terrain relief etc. Anuta (1977) differentiates between
(a) the "open-loop" rectification employing merely information on predictable
geometric errors derived from the imaging process, Sensor attitude and posi-
tion (compare Fig. 1c); and(b) the "fine correction" using ground control
points. Photogrammetrists have been working in this field for many years
(see ISP - Working Group III/1 on "Metric Aspects of Remote Sensing",
operating since 1972).
It is well established that presently available digital satellite images
(Landsat, Nimbus) can be rectified with remaining errors of less than one
pixel and limits set by the geometric resolution (Figure 3). For aircraft
Scanner images, the rectification &ccuracy has been found to be several pixels
(Baker et al., 1975). The limits of accuracy seem to be set by the random
errors of sensor motion &nd imaging, not by geometric resolution. The
accuracy numbers for both the satellite and the aircraft strongly depend
on the density of ground control. Experiences on the interrelation of image
resolution, type of imagery and control point density are however not ex-
tensive.
Geometric rectification may result merely in a mathematical expression
for the deformations, or it may in addition produce a rectfied image.