The transformation equations are computed anew for each iteration. The GCP with the largest distance between the computed coordinates and the actual coordinates is deactivated and the next iteration is started. Filtering stops when the overall RMS error is below the given threshold or when the number of active GCPs is equal to the required minimum number of points or when the optional maximum number of iterations (option iterations) has been reached. With the -d flag, the largest GCP error is used instead of the overall RMS error.
m.gcp.filter uses by default the results of forward transformations (source to target) for filtering. With the -b flag, the results of backward transformations (target to source) are used.
The status of GCPs (active / inactive) is only updated with the -u flag. GCPs that have been filtered out will be deactivated, not deleted.
The transformations are:
order=1:
e = [E0 E1][1].[1]
[E2 0][e] [n]
n = [N0 N1][1].[1]
[N2 0][e] [n]
e = [E0 E1 E3][1 ] [1 ]
[E2 E4 0][e ].[n ]
[E5 0 0][e²] [n²]
n = [N0 N1 N3][1 ] [1 ]
[N2 N4 0][e ].[n ]
[N5 0 0][e²] [n²]
e = [E0 E1 E3 E6][1 ] [1 ]
[E2 E4 E7 0][e ].[n ]
[E5 E8 0 0][e²] [n²]
[E9 0 0 0][e³] [n³]
n = [N0 N1 N3 N6][1 ] [1 ]
[N2 N4 N7 0][e ].[n ]
[N5 N8 0 0][e²] [n²]
[N9 0 0 0][e³] [n³]
In other words, order=1 and order=2 are equivalent to order=3 with the higher coefficients equal to zero.
Last changed: $Date$