Specifying the CP distance

[Velson Horie]

I have selected a pair of CPs (13) using a magnification of ca 200%. So the error on the CP is unlikely to be >2 pixels.  The initial estimated error, "distance", is 14, which after optimising reduces to 12. A similar CP (1) distance looks also about <2 pixels but is estimated at 24, which optimising does not alter.

In general, the algorithm for choosing CPs appears remarkably accurate. Only rarely do I spot a significant error in position matching.

Can the "true" distance (i.e. error) be imposed on the stitching calculations, which presumably would force a calculation of the underlying distortion of the received "plane"?

This would be similar to the calibration of a sensor's 3D colour space by providing multiple "true" 3D positions into which the measured 3D signals are correlated to enable the intermediate positions to be extrapolated more accurately.

This may require a more complete (?and regular) coverage of the overlap field.

I have added a series of CPs down the stitching line (which moves around after re-optimising), and it would be useful to be able to visualise their positions in the Detail Viewer.

[Erik Krause]

Am 21.07.2024 um 10:31 schrieb Velson Horie:

> I have selected a pair of CPs (13) using a magnification of ca 200%. So the
> error on the CP is unlikely to be >2 pixels. The initial estimated error,
> "distance", is 14, which after optimising reduces to 12. A similar CP (1)
> distance looks also about <2 pixels but is estimated at 24, which
> optimising does not alter.

I'm not sure if I understand you correctly. However, if the CP distance
is larger than it should be, two reasons are possible: Either you have
not allowed enough parameters for optimization, or other control points
are positioned incorrectly. PTGui tries to minimize the distance of all
control points by optimizing all allowed parameters.

--
Erik Krause
http://www.erik-krause.de

[PTGui Support]

PTGui adjusts the project parameters (yaw/pitch/roll, lens parameters,
etc) based on the control points. As such it has a limited degree of
freedom.

Perfectly scanned images would need only shift and rotation, and this
can be done perfectly by PTGui. But at 2000 dpi (or so?), mechanical
imperfections (slip, resonance) in the scanner hardware can easily cause
small errors and these are not modeled (and thus not corrected) by
PTGui. Unfortunately your line drawings are extremely sensitive to
alignment errors.

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com