how does rotation of an image increase calculation time?

[Bill Christens-Barry]

I've only used PTGui a few times, and those occasions were about 10 years ago when image sixes were smaller ;-}.

I have questions about how long registering a set of adjoining images might depend on how differently rotated images in the set may be. My sense is that rotation between overlapping images will have a far greater affect on how long it takes to find good or optimal registrations than would occur if the images were only mispositioned in the absence of rotation.

In our case we'll be working with a 2 x 2 or a 2 x 3 array of images, each single channel image being roughly 100 megapixels in size at 16-bit precision. These images may be mis-rotated relative to their neighbors by probably less than 1 degree. To improve throughput, how much effort should be made during image captures to further reduce mis-rotation between adjoining images?

Related question: If I am working with a 4 x 4 array of images, one approach might be to use an iterative approach involving first treating the 4 x 4 array as 4 arrays each of which is itself a 2 x 2 array, and after each of these subarray registrations have been completed to then carry out registration of these four subarrays. This kind of iterative approach could also be done for other problems with different numbers of subarrays, for example 8 x 8 or 4 x 2 etc.

In a similar vein, if I have an array of separate single channel ( Red, Green, and Blue) images, should registration be done first on all images in the same channel first and then to register the stitched Red, Green, and Blue channel images?

Finally, in the absence of identified control points, might it save time to first do an alignment using a low resolution version of the images to get a crude alignment and to then somehow use what is learned from that to do an alignment of the original, higher resolution images?

TIA,

Bill

[PTGui Support]

Hi Bill,

Rotation of the images doesn't matter. PTGui looks for features in the
images, then creates a rotation-invariant descriptor of those features.
This will match with the same feature in other images, independent of
rotation.

Feature matching is done in greyscale.

You can speed up alignment by first roughly placing the images (using
Align to Grid or using a template). Then in Project Settings - Align
Images behavior, check the box 'Images are already roughly positioned'.
PTGui will only search for matches in neighboring images instead of
across all images.

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com

> --
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/
> ptgui/22ec3a21-ce91-42db-a3a7-a38a31de3264n%40googlegroups.com <https://
> groups.google.com/d/msgid/ptgui/22ec3a21-ce91-42db-a3a7-
> a38a31de3264n%40googlegroups.com?utm_medium=email&utm_source=footer>.