Ellipsoid fitting based on maximum intensity and seeded watershed

[maarzt]

The current ellipsoid fitting does not work well, many meta parameters have default values, which would need to be further investigated. Approaches like star dist could be used instead, but the require NVIDIA graphics cards and installed drivers.

We should try something simpler.

1. For each spot:
2. Calculate the maximum intensity.
3. Crop a large enough part of the image arround the spot.
4. appropriate blurring
5. Do a threshold of the surrounding area. Use halve the max. intensity as threshold.
6. Do a seeded watershed to remove under-segmentation.
7. Treat the resulting segment as a values of uniform distribution function. Calculate mean and variance of this distribution. Scale the covariance matrix by a fixed factor to make it fit when segmenting an ideal ellipsoid.

[stefanhahmann]

@tinevez Could you give me access rights in this repo so that I can make some edits in this issue?

[tinevez]

@stefanhahmann Done!

If you are going to implement a new algorithm, can you keep the first one as well?

[stefanhahmann]

@stefanhahmann Done!

If you are going to implement a new algorithm, can you keep the first one as well?

Thanks for the rights.
And yes sure, we will keep the old for comparison as well.