Topic: Automatic Destriping
The version of COSI-Corr released on Feb 3rd, 2009 offers the possibility for automatic destriping of correlation images. Useful mostly to those of you analyzing ASTER imagery (but not limited to), you have surely noticed undulations in the correlation images, due to uncorrected satellite attitudes (roll, pitch, yaw). Until the last version, you were asked to rotate the correlation so that artifacts to be removed be aligned with the image columns. Visually estimating the rotation angle was tedious and inaccurate. Because the undulations are due to attitude disturbances, they occur in the satellite focal plane. In the new version, we therefore determine an approximate (1st order approximation for a quick algorithm) transformation that back-projects the correlation data in the satellite focal plane. Destriping is then done in this geometry, with artifacts necessarily aligned, by construction, with either the rows or columns of the back-projected image. No more tedious determination of the rotation angle is needed. The correction is determined in this geometry, and re-projected on the ground to be applied to the correlation data. This way, no resampling of the correlation data is needed, and correlation quality is preserved.
A correlation image is influenced, by construction, by two images. For destriping, the correlation image is back-projected to the satellite focal plane, but the question remain, which image of the pair should be selected for the back-projection? In most cases, the 2 images you have correlated will share similar orbit inclination, and all artifacts will share the same orientation. In that case, it does not matter which image information is used for destriping. However, if the correlation image presents striping artifacts that can be attributed to an image in particular, it is then best to use the ancillary data and GCP from this image for destriping. If for example you are correlating SPOT and ASTER images, and you see attitude artifacts that you can attribute to the ASTER image, and CCD artifacts (because we only correct CCD artifacts for SPOT 2 / SPOT 4 - HRV1 instruments) from the SPOT sensor, you may want to destripe the correlation twice, in each across-track and along-track direction, using the ASTER ancillary data to correct for attitude patterns, and using the SPOT ancillary data to correct for the CCD stripes.