Stitching suggestions

[Kenton Arkill]

Hi I'm using fiji and have roughly 50 (transmission electron ) images in
each grid. However, the microscope misses the odd one (no idea why) so the
'grid/collection' stitching does not work as it is missing some but only a
few numbers.
I use the 'stitch directory of images, unknown configuration' but with
50ish images take a long time and makes mistakes.

Is there a suggested work around to use the grid sequence? for example put
in blank images where there is one missing, and if so should I put in a
black, white or mean image?

Thanks for suggestions
Kenton

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

[Majurski, Michael Paul (Fed)]

Kenton,

I find myself plugging this work more than answering support questions, which could either be good or bad depending on how you look at it.

I am not an expert on the grid collection stitching plugin. However, as one of the authors of an alternate stitching tool for microscopy images I can suggest how to best use my tool MIST.

Fiji Update Site:
MIST - http://sites.imagej.net/NIST-ISG-MIST/ 

Project Page:
https://pages.nist.gov/MIST/


This is not as general a tool as Grid Collection stitching, but it can be faster and more accurate when the image grids match the assumptions.
First: it only performs 2D stitching (no volumetric stitching). However, it does handle timeseries but no registration is performed between time points. Second: the grid of images must have approximately constant overlaps for the vertical and horizontal directions. For example, vertical overlaps of 10+-2% and horizontal overlaps of 13+-1% would work. The horizontal and vertical image overlap can be different, but per direction the overlaps need to match.

MIST handles missing image tiles so long as the remaining images form a single 4 connected region (i.e. you cannot have two non-connected sub-regions within the image collection).

MIST has also been designed with speed in mind, so runtime is generally limited  by image read time from disk (the tool can leverage native libraries, and run on the GPU for computation acceleration).



For your specific problem:
I have used MIST to stitching SEM images in the past and found that prefiltering the images with a Gaussian smoothing kernel helps the stitching algorithm find correct translations. In my past experience electron images have a higher level of noise. If your images do contain significant noise, that can impact stitching accuracy. To remedy this, you might want to consider prefiltering the images (Fiji command "Process > Filters > Gaussian Blur". You can filter the images, run the stitching, and then use the positions file to assemble the stitched image from the original images so you have an unmodified stitched image. Without seeing an example of your images I can only guess whether this will help.

The prefiltering might help the grid collection stitching algorithm find stronger links between images, enabling the "unknown configuration" options to work better.

As for adding a blank image in place of the missing, that could be problematic given the phase correlation image alignment algorithm (with normalized cross correlation alignment cost function) at the heart  of the grid collection stitching plugin. When I have to replace missing image tiles for stitching I compute the pixel intensity mean and standard deviation of the rest of the images and then build a synthetic image with pure Gaussian noise with the same parameters (mean, standard deviation). This makes a strong spurious correlation less likely.  

Good luck stitching,
~Michael Majurski


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Kenton Arkill
Sent: Thursday, August 18, 2016 06:39
To: [hidden email]
Subject: Stitching suggestions

Hi I'm using fiji and have roughly 50 (transmission electron ) images in each grid. However, the microscope misses the odd one (no idea why) so the 'grid/collection' stitching does not work as it is missing some but only a few numbers.
I use the 'stitch directory of images, unknown configuration' but with 50ish images take a long time and makes mistakes.

Is there a suggested work around to use the grid sequence? for example put in blank images where there is one missing, and if so should I put in a black, white or mean image?

Thanks for suggestions
Kenton

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

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