coregistration is based on the idea that random points are distributed over the image and local statistics are extracted at these points for each image (based on wht window size in azimuth and range). Then the window is slightly shifted to see if the correlation between the reference and the support product is increasing or decreasing. If the correlation is within the GCP tolerance, the point is kept, otherwise it is withdrawn.
This is done iteratively to find points which correspond best between the images to perform an accurate stacking.
Sorry for the late reply and thank you for your explanation
After reading “carefully” the Help Contents, I understood that GCP Tolerance is the limit (must lower than this value) for “row or column shift of the new slave GCP” in coarse registration.
Are the words “reference” and “support product” in your description mean “master” and “slave”, respectively?
However, I am still not quite sure about the Fine Accuracy in Azimuth and Fine Accuracy in Range.
Can you provide me more explanation?
yes, we want to replace these two terms sucessively, which of course might cause some confusion during the time of transition. This has been suggested and approved here.
I’m not 100% sure about these parameters, maybe someone else can comment?