The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1358
2. STEREO ORTHOKIT DATA PRODUCTS
Data products from the Fore and Aft sensors of Cartosat-1 are
being operationally generated and disseminated to users from
the National Remote Sensing Agency (NRSA), Hyderabad,
India among others. Out of the many levels of products, the
following three types are especially suited for the commercially-
off-the-shelf (COTS) available photogrammetric software
packages to support.
1. Standard Mono Products
2. Orthokit Mono Products
3. Orthokit Stereo Products
The orthokit stereo products are described in little more detail in
the following subsection. Image data is by default in GeoTIFF
format based on the Tagged Image File Format (TIFF) which is
widely used today. This format is supported by all commercial
software and is therefore easy to integrate. The Geographic
extension (Geo) part of the format is supported by all
photogrammetric software packages.
The Geo part of GeoTIFF basically adds geo-referencing
information from the image file to the TIFF file (geographic
coordinates of the top-left comer and pixel sizes) and may also
specify the map projection and geodetic system.
2.1 Orthokit Stereo Products
contain a pair of radiometrically corrected 10-bit image files
corresponding to the Fore and Aft sensors in GeoTIFF format,
two (encrypted and decrypted) ASCII files containing the
rational polynomial coefficients (RPCs) respectively for each
image, and metadata files in ASCII format. These products
enable further processing to generate an orthoimage by way of
improving the ground-to-image relationships as provided in the
RPC files using ground control points (GCPs).
Orthokit products are geometrically uncorrected except for the
tagging of centre and comer coordinates with system corrected
geographic latitude-longitude values. However detailed ground-
to-image relationship(s) is (are) explicitly provided in the RPC
file(s).
3. ANOMALIES REPORTED
More than two investigators consistently reported that Fore
images supplied were somewhat blurred as compared to the Aft
images, even while investigating different test sites. In certain
cases, it was observed that the rational polynomial coefficients
supplied were having the divide-by-zero anomaly resulting in
inconsistencies in geometric processing. Also, in one of the test
sites, the 0-360 degrees longitude convention adopted was
found to result in software failures at user end. These anomalies
have since been rectified. In the case of geometric anomalies, it
was necessary to rectify the problem and provide corrected data
sets to all users including those who faced these problems on a
quick turn-around-time basis. The radiometric anomaly was not
fixed immediately as it required a long lead time. Subsequent
sections describe how these improvements were incorporated.
4. IMPROVEMENTS IN RADIOMETRY
Improvements in radiometry could be realised primarily based
on the improved point spread function estimates provided by the
payload design team from Space Applications Centre, ISRO for
both the Fore and Aft sensors, subsequent to the C-SAP launch.
However, the improvement process was not straight forward, as
our attempts to improve radiometry, in addition to enhancing
the basic image signals also enhanced the inherent noises
present in the image signal.
4.1 Restoration & Denoising:
On analysing the specific phase-I data sets as well as other
phase-II data sets of C-SAP test sites, the major source of noise
signals were observed to be due to the residual uncorrected
stagger component, especially in Fore images. Stagger
correction is incorporated primarily to take care of the five
pixels shift between odd and even detector elements in the focal
plane of the CCD (charge-coupled device) detector arrays used
in the Fore and Aft sensors. It has been observed that in certain
cases, the standard algorithm for stagger correction for stitching
the odd and even pixels of a scanline together to form a single
full scanline was not performing adequately.
4.1.1 Scene-based stagger removal:
The stagger value between odd and even pixels changes scene
to scene depending on the latitude and the variation of the
dynamic component of the orbit and attitude values (like rate
component). During the data pre processing (radiometric
correction), the stagger value is computed using the ephemeris
and is used for the alignment of the pixels. The sampling
interval (125 ms) for attitude and orbital elements in the
ancillary information is not sufficient to characterize the
satellite behavior for smaller intervals of time such as stagger
induced time difference for acquiring odd and even pixels of a
scanline (5 scans or 1.83 ms). During the careful inspection of
radiometric corrected image at the larger scales, the residual
stagger effects can be seen to an extent for which small
additional correction is required. This residual stagger must
have been due to the inadequate sampling interval of the attitude
information.
The scene-based estimation of stagger can be done using
relative control points (RCP) between the odd and even images.
RCPs can be derived using digital cross correlation based image
matching technique between the odd and even images.
First the data set is prepared for correlated RCPs at every scan
line. The differences in odd and even data are supposed to be
the stagger in the raw image. A histogram approach is adopted
to quantify the stagger number in along track direction for one
thousand samples. The RCPs are generated at 12 positions in
across track direction viz. pixel numbers 50, 500, 1000, 1500,
2000, 2500, 3000, 3500, 4000, 4500, 5000, and 5500 with one
scan sampling interval in along track direction.
4.1.2 Candidate artifact zones identification &
preprocessing:
As the MTF correction process or restoration behaves in a
peculiar way at high reflectance regions and high gradient
regions in the form of ringing effects (alternate dark and bright
lines) and false shadows (dark lines/ spots), these candidate
artifact zones are first identified and locally pre-processed so
that after restoration, no such artifacts are really present.
Pre-processing involves detection and smoothing of high
reflectance objects prior to the process of restoration. Automatic
detection of bright objects in the original radiometrically
corrected image is achieved by a tree-based segmentation
approach. An image typically consists of several objects
(features) placed over slowly varying background. In the