Page 1 of 1
Aligning images
Posted: July 25th, 2013, 6:38 am
by JerryF
I am having dificulty in aligning handheld images for use in the HDR/Stack transform.
Lining up two is not dificult in "Composite" but of course they forget the alignment when I open HDR/Stack.
Looking through previous topics there seems to be no easy answer except to use an external program.
Is it any better with PW7? I am still on PW5P but thinking of investing sometime in the next few months.
Has anyone worked out a good work around?
thanks
Jerry
Re: Aligning images
Posted: July 25th, 2013, 11:01 am
by jsachs
If you set the operation in Composite to "Register" and perform the necessary alignment, the result image will be a version of the overlay image aligned with the input image. You can then use this aligned image to stack with the base image or with any other image aligned to the base image.
Re: Aligning images
Posted: July 30th, 2013, 2:24 pm
by Robert Schleif
It would be nice to be able to use the "refine" button in the Composite-Register for automatic refinement of fairly closely aligned images. The automatic refine seems not to work, however, when the densities of the two images differ significantly, as they do in the images prepared for HDR. (I could get it to work on two copies of an image in which the intensity of one had been multiplied by 0.8, but it would not work when the multiplication was 0.6.)
Re: Aligning images
Posted: July 30th, 2013, 4:12 pm
by jsachs
I will take a look at aligning high pass versions of the base and overlay images which should largely eliminate the effects of differing brightness -- assuming of course that detail around each alignment point is not lost due to clipped shadows or highlights.
Re: Aligning images
Posted: August 29th, 2013, 9:16 am
by Robert Schleif
Does the high pass method work? I wonder if the amplitudes of the higher frequency components, as well as the lower frequency components also differ by the exposure difference between the two images and therefore that the method might not work. Might a general approach be to multiply the components from one of the images by a series of numbers, i.e. 20, 14, 9.8, .....1, 0.7, 0.49...0.05 and for each of these, seek correlation with the components from the other image? Although this would run as much as 20 x slower than the current algorithm, it still might be fast enough.