Alignment

[Guest]

When a piece of software aligns a set of images is it able to do this in small sections across an image or is it only done by shifting and rotating the whole image?

[vlaiv]

1 hour ago, woodblock said:

When a piece of software aligns a set of images is it able to do this in small sections across an image or is it only done by shifting and rotating the whole image?

Can you be more specific - explain what you mean?

It usually matches two images by having single reference image and other image that will be transformed to match reference image. Often you can choose what type of transform is "allowed".

If you want to match two sections across the single image then you can do this with above method if you for example - make copies of selections of that single image and treat them as two separate images.

If you want to do some sort of autocorrelation (finding transform of original image that would match it to itself in particular zone or something like that) - then I don't think you'll find such feature in general purpose software. Your best bet is that someone already made specific piece of software that does exactly what you want or to try to emulate your self with set of operations - or perhaps to write your own using macro language of some package (like Mathematica or Matlab or something like that).

[Guest]

Previously I had assumed that aligning one frame to another meant that the second image was shifted and rotated as a whole to match the reference frame. Like you would do manually. If you had two printed images you could lie one on top of the other and shift it about so that the images matched. So the aligned image was moved 'as a whole'.

But I thought about it when taking pictures of the surface of the moon last night. It was very clear and I had a sharp image but I could see the image rippling because of the atmosphere moving about. But it doesn't move as one. Parts of the picture moves one way and other parts move another way. Flipping through the pictures this morning they look quite sharp but I can see that surface features are shifting about from one frame to the next. So I wondered if the alignment routines have some clever way of aligning parts of the image independently. So in order to perfectly align one image against a reference image it would have to shift this part of the image this way and another part of the image a different way. Obviously it would have to kind of distort things a bit so that it all 'connected up'.

 

[vlaiv]

Ah I see.

Yes, but that is different type of alignment process. Regular alignment process that you described aligns whole image in the way you described and often uses affine transform to do so and rarely does non linear correction in order to correct for lens distortion on wide field images.

There is another type of alignment process used in planetary imaging that works differently. It uses alignment points. Each point in one frame is checked against number of neighboring points in other frame and correlation is calculated (match is performed). This creates list of "same" / matching points across the frames.

True position is calculated as mean geometrical point across the frames and then each frame is aligned against that calculated reference frame (which is just list of geometrical positions where each alignment point should be placed). Then unwarping of each frame is done so that its alignment points match reference frame and then stacking is performed.

So yes, you are right - here frames are warped or rather unwarped when aligning.