n
se
1€
of
in
« 4 OQ
V4
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Er tte
Input
Images
Balancing of
Grey Level
—
c |
D
Classification
M orphological
Pixel/Patch
and Mask Filter Ranking
C loud -free M osaic M erging of
M osaics
Production
Sub-images
Figure l. A schematic diagram of the cloud-free and cloud shadow-free mosaic generating system
The initial cloud mask produced using a fixed intensity
threshold usually excludes the thin clouds at cloud edges. To
solve this problem, a morphological filter is applied to dilate the
initial cloud mask patch so that the thin clouds at cloud edges
are included in the cloud mask.
2.4 Pixel/Patch Ranking
In the masking process, some bright patches on bare soil can
still be mistaken as cloud patches due to the similar intensity
level and colour. These patches can be unfilled because no good
data patches in the set of co-registered cloudy images can be
used to mosaic the composite image. We therefore introduce a
pixel/patch ranking procedure that employs the pixel/patch
intensity and some suitably chosen band ratios to rank the
pixels/patches in the order of "cloudness" and “shadowiness”
according to some predefined ranking criteria described below.
In this procedure, a shadow intensity threshold 7, a vegetation
intensity threshold 7,, à cloud threshold 7, are determined from
the intensity histogram. The pixel/patch ranking procedure uses
these shadow and cloud thresholds to rank the pixels/patches in
order of *cloudiness" and “shadowness”. Each of the non-cloud
and non-shadow pixels/patches in the images is classified into
one of three broad classes based on the band ratios: vegetation,
open land and others.
For each image n from the set of N acquired images, each
pixel/patch at a location (i, /) is assigned a rank r,(i, /) based on
the pixel/patch intensity Y,(/, /) according to the following
rules:
^ar , 7 7 < (4 V SY + since =
D For 7,.$ (Yo "95$ TOW fr, Y, and class
"
» stafi " » . Cp +
vegetation", then ry, € ry;
2) Por 7,50
€ " a ^ M y +
land", then r,, € rj;
3 Cy « € / E ^ ey
3) Hy 7, and Yı, > Te; then r, rye
S | V y a i£ V ~ J » ‘ «p
4) ForY,» J^ Tof p" then rj; <r;
5) For!
S ForY,,
ups T if Y, Y, and class open
Y, > le ity, < Y then ry, <r;
In this scheme, pixels/patches with lower rank values of r, are
more superior and are more likely to be selected. Pixels/patches
with intensities falling between the shadow and cloud
thresholds are the most superior, and are regarded as the "good
pixels/patches". The “good pixels/patches” are further classified
into “vegetation pixels/patches” or “open land pixels/patches”
depending on whether the pixel/patches intensity is below or
above the vegetation threshold. As rule of thumb, the darker
*good pixels/patches" are preferred over the brighter "good
pixels/patches" because the brighter "good pixels/patches" may
be contaminated by thin clouds. Where no good pixels/patches
are available, the "shadow pixels/patches" are preferred over the
"cloud pixels/patches". Where all pixels/patches at a given
location are "shadow pixels/patches", the brightest shadow
pixels/patches will be chosen. In locations where all
pixels/patches have been classified as "cloud pixels/patches",
the darkest cloud pixels/patches will be selected. After ranking
the pixels/patches, the rank-r index map ni, /) representing the
index n of the image with rank r at the pixel/patch location (7, /)
can be generated. It is preferred that only the rank-1 and rank-2
index maps are generated and kept for use in generating the
cloud-free and cloud-shadow free mosaics.
2.5 Merging of Sub-images
The rank-1 and rank-2 index maps generated from the
pixel/patch ranking procedure are used to merge the input
multi-scenes that have been processed by the grey-level
balance. If the pixel at a given location has been classified as
“vegetation pixel”, the pixels from the rank-1 image and the
rank-2 image at that location may be averaged together in order
to avoid sudden spatial discontinuities in the final mosaic
image. Otherwise, the pixels from the rank-1 image are used.
2.6 Mosaic Production
The basic idea of the cloud-free and cloud shadow-free mosaic
algorithm is to mask the clouds, cloud-shadows and select good
image data from the different scenes in a selected set of co-
e
Un