Page 1 of 1

Noisy surfaces, fine details, old graffiti

Posted: Tue Nov 14, 2017 9:03 am
by paulwright
Hi all

I have been working on this point cloud for some time, with little positive outcome. Before I give up, I wondered whether someone more skilled, from the community, might have any advice?!

I have been asked to scan an internal wall of an old house (Some of the data can be found as an ASCII file here: http://bit.ly/2zFeKOc ). The task is to document old graffiti (in this case scratches/carvings in plaster rather than spray paint!), and I am really struggling to see anything clear in the data.

In the attached file, you will see a deeply etched set of initials (RL), on well dressed stone (presumably a Masons' Mark), in the bottom left. That is clearly visible in the cloud and any resultant mesh.

There are, however, a number of etchings to the right of this - easy ones to spot with the naked eye on site are the circle near the bottom of the scan (interpreted as a Witches Mark), and two shields with diamonds and chevrons towards the middle right. You can JUST make these out in the cloud (if you can get the shadows right), but they are almost non existent in any mesh at any detail. These carvings are more delicate and made in plaster. This surface is already rougher than the stone.

So is this just an issue of the surface being 'noisier' than the detail I am trying to pick out, or are there any hints to try things that might make the detail 'pop'?

It may also be that the technology I used is not up to the task. This is a finding in itself, as it informs the curator's methodology. However, before I blame the kit, I wanted to just check that I had exhausted all possibilities with the processing end.

Thanks

Paul

Re: Noisy surfaces, fine details, old graffiti

Posted: Tue Nov 14, 2017 1:31 pm
by deepminder
Hi Paul,

I can clearly see the "RL" and some other letters on the left part of the pointcloud.
Image
But I can't see anything except some scratches on the right part of the pointcloud. Can you make some screenshot with markings, so we know what to look for?

Regards,
Denys

Re: Noisy surfaces, fine details, old graffiti

Posted: Tue Nov 14, 2017 3:21 pm
by daniel
If it's only for display, you can use the normals with the custom light:
cc_normals_with_custom_light.JPG
cc_normals_with_custom_light.JPG (81.8 KiB) Viewed 5001 times
Simply compute the normals with a not too big radius (so as to capture the small details - here I chose 0.005) and with the right direction (as you have a flat wall).

Then:
- toggle the light off (F6)
- toggle the custom light on (F7)
- find it (it's a yellow cross, that can be quite far from the cloud sometimes ;) and bring it closer to the cloud with CTRL + right mouse pressed. It may be wiser to do it before computing the normals as display with normals is quite slow.
- position it so as to obtain a raking light

I believe another way would be to use PCV, but it may depend on the noise level. I'll try that on my side.

Re: Noisy surfaces, fine details, old graffiti

Posted: Tue Nov 14, 2017 3:58 pm
by daniel
Here is the result with PCV:
cc_pcv.JPG
cc_pcv.JPG (241.23 KiB) Viewed 4999 times
Make sure the (wall) plane is aligned with XY (you can use the Level tool). Then use PCV with a big resolution (e.g. 2048) and a lot of rays (e.g. 1000). Then increase the saturation of the resulting scalar field.

It may not be as clear as the previous view with the normals (because of the noise) but at least you don't have to play with the custom light position.

Last but not least you can also use the hillshade option of the Rasterize tool:
cc_rasterize.jpg
cc_rasterize.jpg (220.01 KiB) Viewed 4999 times
Once again it would be better to set the plane (wall) parallel to XY or YZ. Then use a not too small grid step (e.g. 0.002) to smooth the cloud a little bit.

And then generate a 'hillshade' layer with a raking light (> 70 degrees). You can play with the azimuth angle to enhance the carvings depending on their orientations. You can also export the resulting grid as a cloud (with the hillshade values as a scalar field). And you even once again enhance this scalar field saturation.