TrakEM2: Alignment of serial sections (CNS)


I want to align 26 individual images, each measuring 1280x960 pixel, section thickness of 60 nm. The pixelsize
is about 30 nm.
My approach was: (modified from:!topic/fiji-devel/FET8-Cs-qRg)

  1. Adjust>Image Filters `Robust Local Contrast: Standard parameters
  2. Least square alignment, rigide: Standard parameters
  3. Least square alignment, affine: Standard parameters
  4. Elastic alignment, rigide:
    Scale of 0.5 (because of 30 nm pixelsize/ 60 nm section thickness)
    Search radius 200
    Block radius 100 (or should this be 200 too?)
    Resolution of 35
    Test 5 layers
    (layers are pre-aligned checked)

The final, elastic alignment took about 5-7 hours (over night) but drastically improved the previous alignment.

The result of this workflow was “ok”, but there is still a significant moving between the individual sections.
I already used test blockmatching parameters, but I have problems to read the final image and interprete the colours.

Best regards,

Sorry to be dense, but… what is the concrete question? How to make the alignment more precise?


yes, sorry for the vague question…But I could offer the raw test-data and my result (video). I tested some parameters for elastic alignment (scale, resolution, image radius), but there were still some “jumps” in the stack.

So my plan is to make the alignment more precise. Unfortunately, the compression artifacts between the single sections/images vary a bit, because of minimal differences in the section thickness. I think, this could be a big problem.

Hi @CaDi,

If you havn’t already, read this page which has the information you need about tuning various parameters. Look here for information on interpreting Test Block Matching.

If its still unclear, we can perhaps give more concrete advice if you are willing to post the output of Test Block Matching Parameters.

Good luck!

1 Like

Hi John,

thanks for your reply. I already checked these pages, but the results were not good. I’ll prepare some more detailed steps of my workflow and post the Test Block Matching Parameters.
I think, the sample is one source of the errors. Using other raw data, I had almost perfect alignments with no visible distortions in reconstructed X/Y section planes using 3Dmod.

Best regards,