Alexandre Neto <[email protected]> writes: > So, I think you need to sum the squares of each GCP residual, divide it by > the number of GCP, and then get the Square root of the result.
That makes sense, but there's a deeper question about what the OP is trying to calculate. Beware that the following details are out on a limb: There are 4 degrees of freedom for georeferencing if one assumes constant scale of the source image, basically an x/y point, a scale, and a rotation. So with 2 GCPs you will get zero residuals. But that doesn't mean your image is perfectly aligned. If you are trying to estimate the RMS error for points that aren't your GCPs -- which I think you might be, then you might want to subtract from the GCP number to account for errors absorbed by estimated parameters. Basically, dividing sum of squares by GCP-2, for four estimated parameters. If you have lot of GCPs, this starts to not really matter. If you really care about characterizing the error, I'd suggest having at least 10 and maybe 20, without any real basis for those numbers. Here's an example of language about RMS (search in page) https://docs.digital.mass.gov/dataset/massgis-data-usgs-color-ortho-imagery-2019 Note that this is "we checked 132 points", not "we used 132 points to align", so there one should divide by the GCP number - they really are measuring errors in the dataset at a number of points, not looking at residuals.
signature.asc
Description: PGP signature
_______________________________________________ Qgis-user mailing list [email protected] List info: https://lists.osgeo.org/mailman/listinfo/qgis-user Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-user
