problem with grid/collection stitching

[Aryeh Weiss]

I am having trouble with stitching the following set of images (filename
defined positions):

https://drive.google.com/file/d/0Bytu294eIohjUE5waC1EQUhfZE0/view?usp=sharing
The above link is to a zip file that contains the 120 tiles that I am
trying to stich (nominally 10% overlap)

With the default parameters, the following is obtained, with many
"misplaced" tiles
Here is a link to what I get.
https://drive.google.com/file/d/0Bytu294eIohjbzYwWG15bW5qWEk/view?usp=sharing

If I reduced the regression threshold to 0.02 (from 0.3), it is better,
but I still have the (1,1) tile misplaced:
https://drive.google.com/file/d/0Bytu294eIohjbDQ0N2VxVVVuUVE/view?usp=sharing

If I repeat with the smaller regression threshold, but I only stitch the
top half of the image, including the (1,1) tile, it works correctly.

By eye it looks like I have enough  "structure" (marks, spots, etc) to
align. In any case, I dont see how it can get so misalligned, even when
I reduced the max/avg displacement.

Is there something inherently wrong with  with my images? Is there a
better set of parameters which I can use to get it right.

Thanks in advance for any help you can provide.

--aryeh

--
Aryeh Weiss
Faculty of Engineering
Bar Ilan University
Ramat Gan 52900 Israel

Ph:  972-3-5317638
FAX: [phone]

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Jan Eglinger]

Dear Aryeh,

I didn't have time to check out your images yet, but had some problems
with misalignment in the past as well. See also this discussion on the
ImageJ forum:
http://forum.imagej.net/t/limit-the-tile-displacement-in-stitching/424?u=imagejan

If I understood correctly after looking at the source code, the max/avg
displacement parameter defines how far each tile can deviate in the
global optimization from its computed position in the pairwise alignment
with each overlapping tile. That means if the pairwise alignment
returned "wrong" values, the tile will get misaligned no matter what are
its coordinates defined in the TileConfiguration.txt.

Maybe Stephan Preibisch can comment more on that.

Best regards,
Jan


On 24.08.2016 05:49, Aryeh Weiss wrote:

> I am having trouble with stitching the following set of images (filename
> defined positions):
>
> https://drive.google.com/file/d/0Bytu294eIohjUE5waC1EQUhfZE0/view?usp=sharing
>
> The above link is to a zip file that contains the 120 tiles that I am
> trying to stich (nominally 10% overlap)
>
> With the default parameters, the following is obtained, with many
> "misplaced" tiles
> Here is a link to what I get.
> https://drive.google.com/file/d/0Bytu294eIohjbzYwWG15bW5qWEk/view?usp=sharing
>
>
> If I reduced the regression threshold to 0.02 (from 0.3), it is better,
> but I still have the (1,1) tile misplaced:
> https://drive.google.com/file/d/0Bytu294eIohjbDQ0N2VxVVVuUVE/view?usp=sharing
>
>
> If I repeat with the smaller regression threshold, but I only stitch the
> top half of the image, including the (1,1) tile, it works correctly.
>
> By eye it looks like I have enough  "structure" (marks, spots, etc) to
> align. In any case, I dont see how it can get so misalligned, even when
> I reduced the max/avg displacement.
>
> Is there something inherently wrong with  with my images? Is there a
> better set of parameters which I can use to get it right.
>
> Thanks in advance for any help you can provide.
>
> --aryeh
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Aryeh Weiss]

Dear Jan,

Thank you for your reply.
I suspect that because my image are sparse, I need to reduce the
regression threshold. However, I still dont understand why it works when
I do 1/2 of the image at a time. When it works, even with the very low
regression threshold, it works very well. I examined many of the seams
which I thought would be troublesome, and the stitching was perfect.

So, I stitched the two halves separately, and these are the results:
Top half:
https://drive.google.com/file/d/0Bytu294eIohjUWZhY3JkZDJLQVk/view?usp=sharing
bottom half:
https://drive.google.com/file/d/0Bytu294eIohjSEZhejZ1ekRtSjA/view?usp=sharing

But I cannot get these two images to pairwise stitch, apparently because
of the black borders (my scope software staggers the fields when I
collect with overlap -- I wish I knew how to tell NIS elements not to do
that).

I thought checking "ignore zero" would help, and I set up ROIs in the
areas where I wanted the program to check, but still no success.

  Best regards
--aryeh


On 24/08/2016 10:57 AM, Jan Eglinger wrote:

> Dear Aryeh,
>
> I didn't have time to check out your images yet, but had some problems
> with misalignment in the past as well. See also this discussion on the
> ImageJ forum:
> http://forum.imagej.net/t/limit-the-tile-displacement-in-stitching/424?u=imagejan 
>
>
> If I understood correctly after looking at the source code, the
> max/avg displacement parameter defines how far each tile can deviate
> in the global optimization from its computed position in the pairwise
> alignment with each overlapping tile. That means if the pairwise
> alignment returned "wrong" values, the tile will get misaligned no
> matter what are its coordinates defined in the TileConfiguration.txt.
>
> Maybe Stephan Preibisch can comment more on that.
>
> Best regards,
> Jan
>
>
> On 24.08.2016 05:49, Aryeh Weiss wrote:
>> I am having trouble with stitching the following set of images (filename
>> defined positions):
>>
>> https://drive.google.com/file/d/0Bytu294eIohjUE5waC1EQUhfZE0/view?usp=sharing 
>>
>>
>> The above link is to a zip file that contains the 120 tiles that I am
>> trying to stich (nominally 10% overlap)
>>
>> With the default parameters, the following is obtained, with many
>> "misplaced" tiles
>> Here is a link to what I get.
>> https://drive.google.com/file/d/0Bytu294eIohjbzYwWG15bW5qWEk/view?usp=sharing 
>>
>>
>>
>> If I reduced the regression threshold to 0.02 (from 0.3), it is better,
>> but I still have the (1,1) tile misplaced:
>> https://drive.google.com/file/d/0Bytu294eIohjbDQ0N2VxVVVuUVE/view?usp=sharing 
>>
>>
>>
>> If I repeat with the smaller regression threshold, but I only stitch the
>> top half of the image, including the (1,1) tile, it works correctly.
>>
>> By eye it looks like I have enough  "structure" (marks, spots, etc) to
>> align. In any case, I dont see how it can get so misalligned, even when
>> I reduced the max/avg displacement.
>>
>> Is there something inherently wrong with  with my images? Is there a
>> better set of parameters which I can use to get it right.
>>
>> Thanks in advance for any help you can provide.
>>
>> --aryeh
>>
>

--
Aryeh Weiss
Faculty of Engineering
Bar Ilan University
Ramat Gan 52900 Israel

Ph:  972-3-5317638
FAX: [phone]

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html