Elastic layer alignment in TrakEM2 - problem with correspondences

[Klaus Griebel]

Hello everybody,

my name is Klaus, I'm a PhD student from Erlangen (Germany) and am currently working on reconstructing a TEM series. The picture aquisition is not complete yet, but I already have pictures of 108 slices with roughly 8 pictures per slice. I successfully entered about 60 layers by now, montaged them (elastically) and got decent results for each layer per se.

I encountered problems with the automatic alignment of layers though. So far I have tried both linear and elastic alignments for subsets of these 60 layers with varying numbers of layers compared simultaneously (3-10), varying sizes for feature extraction (min 50-2500; max 600-5000), varying search and block radii (500-2000), varying alignment errors (50-400) but so far no luck.

The biggest problem that I can see is that my layers are always spaced more than usual because I only captured images from every fifth slice or so (with slight variations), due to the immense size of the project (I presum it to end at roughly 200 layers in the end). Therefore I believe the similarities from two adjacent layers are too little for the standard alignment parameters. I believe that I need to define more global features and set a different error value still.

Can anyone help me with the tuning of this process or tell me that my series can or cannot be aligned automatically from what I've posted her?

On a side note, the individual images were taken at an 3150x magnification and are all LZW compressed tif files with a resolution of 4008x2672. The canvas size I got from the autosize function at the moment is 18983x8910 (if that helps in any way). Also I'd like to note that I am quite the novice at this.

Thank you in advance.

Greetings,
Klaus

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

[Saalfeld, Stephan]

Hi Klaus,

how big are your pixels (um per pixel) and what is the physical spacing
between your slices?

Also read this page:

http://fiji.sc/Test_Block_Matching_Parameters

Thanks,
Stephan



On Thu, 2015-06-25 at 06:00 -0400, Klaus Griebel wrote:

> Hello everybody,
>
> my name is Klaus, I'm a PhD student from Erlangen (Germany) and am currently working on reconstructing a TEM series. The picture aquisition is not complete yet, but I already have pictures of 108 slices with roughly 8 pictures per slice. I successfully entered about 60 layers by now, montaged them (elastically) and got decent results for each layer per se.
>
> I encountered problems with the automatic alignment of layers though. So far I have tried both linear and elastic alignments for subsets of these 60 layers with varying numbers of layers compared simultaneously (3-10), varying sizes for feature extraction (min 50-2500; max 600-5000), varying search and block radii (500-2000), varying alignment errors (50-400) but so far no luck.
>
> The biggest problem that I can see is that my layers are always spaced more than usual because I only captured images from every fifth slice or so (with slight variations), due to the immense size of the project (I presum it to end at roughly 200 layers in the end). Therefore I believe the similarities from two adjacent layers are too little for the standard alignment parameters. I believe that I need to define more global features and set a different error value still.
>
> Can anyone help me with the tuning of this process or tell me that my series can or cannot be aligned automatically from what I've posted her?
>
> On a side note, the individual images were taken at an 3150x magnification and are all LZW compressed tif files with a resolution of 4008x2672. The canvas size I got from the autosize function at the moment is 18983x8910 (if that helps in any way). Also I'd like to note that I am quite the novice at this.
>
> Thank you in advance.
>
> Greetings,
> Klaus
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Klaus Griebel]

Hi Stephan,

I haven't calibrated the series yet. I thought it would be nicer/easier to do after I've put in all layers I want and pre-aligned them.

I measured the pixel size though, it is at 0.012738854µm/px-edge (i.e. 1px= 0.012738854µm x 0.012738854µm). Also I haven't typed in all layer distances yet, as we've encounterd "problems" during the cutting process. Basically there was a mixup and I have to determine the distance for every section anew, which I haven't done yet.

So are you indicating that the elastic alignment can't/won't work without the spacing data? So far I've evenly spaced the layers by 0.25µm as a "placeholder".

I have actually found the Test Block Matching Parameter Plugin, but so far have been a little overwhelmed by what to do to get it to work, but it appears I need to first pre-align my series and that I had to do by hand so far. Even the linear layer alignment did not work properly so far, hence I bet I'll have to calibrate, don't I :)

Thanks,
Klaus

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

[Saalfeld, Stephan]

No---you only need a rough estimate of what the ratio between z-spacing
and xy-resolution is.  Starting from there, you can do educated search
in parameter space which usually means trying 10--20 things---not too
bad.  Without even a rough idea that becomes Kaffeesatzleserei.

Cheers,
Stephan



On Fri, 2015-06-26 at 04:24 -0400, Klaus Griebel wrote:

> Hi Stephan,
>
> I haven't calibrated the series yet. I thought it would be nicer/easier to do after I've put in all layers I want and pre-aligned them.
>
> I measured the pixel size though, it is at 0.012738854µm/px-edge (i.e. 1px= 0.012738854µm x 0.012738854µm). Also I haven't typed in all layer distances yet, as we've encounterd "problems" during the cutting process. Basically there was a mixup and I have to determine the distance for every section anew, which I haven't done yet.
>
> So are you indicating that the elastic alignment can't/won't work without the spacing data? So far I've evenly spaced the layers by 0.25µm as a "placeholder".
>
> I have actually found the Test Block Matching Parameter Plugin, but so far have been a little overwhelmed by what to do to get it to work, but it appears I need to first pre-align my series and that I had to do by hand so far. Even the linear layer alignment did not work properly so far, hence I bet I'll have to calibrate, don't I :)
>
> Thanks,
> Klaus
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Klaus Griebel]

Okay I get that. Thank you Stephan.

So, what would be the most efficient way to go about this then? I'm thinking of pre-aligining the stack manually (as registering with least squares didn't work either), then calibrating it and putting in the actual spacing data, then export it as RGB in green and let the Test Block Matching Parameter Plugin run to check parameters?

If there is a way to easier "stream-line" this process I'd be glad to hear them :)

Thanks for the help so far and in advance.

Klaus

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