Noisy hillshade and DEM

[Vertesh Roni]

Hi,

I am trying to generate clean DEM and hillshade for small lunar craters using LRO NAC images, but the final results are noisy. I would appreciate advice on how to improve the workflow and SfS/DEM parameters.
My current workflow is:

1. Download about 10+ NAC images covering a specific crater
2. Run spiceinit
3. Crop the images around the crater
4. Run stereo on several pairs of cropped .cub files
5. Run point2dem for all pairs’ stereo
6. Screen the best DEMs based on comparison to LOLA profiles, using NRMSE
7. Merge the best DEM(s)
8. Run sfs on the merged DEM

The main issue is that both the merged DEM and the final sfs DEM/hillshade are noisy (attached), with no visible improvement, even when the alignment with lola is good.

Thanks

[Oleg Alexandrov]

For a given stereo pair it is suggested to:

 - Check if the stereo convergence angle is say between 10 and 40 degrees. Otherwise results will degrade.

 - Check if the input images have similar illumination. For that, can mapproject both onto your DEM and overlay them. If shadows move a lot, results will be noisy.

 - It is suggested in either case to perhaps blur these a little, then mapproject and redo stereo with the asp_mgm algorithm and subpixel-mode 9. See example here: https://stereopipeline.readthedocs.io/en/latest/next_steps.html#example-for-isis-images

In either case, for SfS, this may do. The noise seems to be local. You can blur it a little, till the noise goes away, such as with dem_mosaic (see its manual for blur sigma option), and run SfS on top.

For SfS, it is strongly suggested to before-hand maproject all images and cameras onto your input blurred DEM. These should agree when overlaid with georeference information except for shadow direction. Diverse illumination is very important. So for stereo same illuminatinon is important, but for SfS need to have different ones in at least some of the images. Choosing the smoothness weight for SfS is also important. For doing SfS at 1 m/pixel we use a smoothness weight of 0.08 or so, but you can experiment with double and half of this and other fractions, and observe how the results change.

We have more details here: https://stereopipeline.readthedocs.io/en/latest/sfs_usage.html

There you will see simple examples early on, and lengthier very detailed example much later in the doc.

We did this process many times before, and I think it should work. You are welcome to reach out offline if having issues.

Oleg

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/79cebcf0-0ff7-4ce8-aa27-7e9370e138d5n%40googlegroups.com.