high frequency noise reduction

[FoodMan]

Hello...

That would be ultra nice if we had some sort of noise reduction tool... currently I use Ultra high recon., while having some amazing results, I do have some high frequency noise in the mesh...

I also have a question... does using more images reduces that noise...?

Thanks
f/

[Wishgranter]

yes, have tested it on my own data.. What EFFECTIVE OVERLAP you get now  ?

 Question is have you a clean low noise imput ? because it can be reduced significantly......

[FoodMan]

hello...

It is a Turntable object... I do have 128 images, but I could only use 8 images for now (waiting a new PC to be able to do it full) .. I was just wondering is having more images would reduce the noise..

my input images are super clean, I shot them in a studio with a canon 5D at f22... (5616 x 3744) so I have almost noiseless photos...

btw, I was reading Meshlab has a Noise reduction tool... I have to test it..

Thanks Wishgranter..

f/

[tweezlednutball]

having more photos/points would create less noise.  Also, we use Neat Video Noise Reduction tool if our photos are noisy which may produce noise in your mesh.  from there we use photoshop batch mode to reinject any EXIF data back into the images.

[Infinite]

having more photos/points would create less noise.  Also, we use Neat Video Noise Reduction tool if our photos are noisy which may produce noise in your mesh.  from there we use photoshop batch mode to reinject any EXIF data back into the images.

I can't imagine a noise reduction tool being of any use as surely it will make each photograph different than what was captured in reality? meaning less chance for the software to find correspondence?

my input images are super clean, I shot them in a studio with a canon 5D at f22... (5616 x 3744)

I would say be careful using high f-stops, anything over 12 or so and you start to get diffraction which reduces image quality. It might seem like a good idea, as you get more control over DOF at different distances but it damages image quality.

[FoodMan]

hello..

Thanks for the advice Lee.. 

[Nico]

+1

Surface-preserving smoothing would be great addition to photoscan. Meshlab has a good library of smooth filters to look at (laplacian surface preserve, hc, two step, taubin). Taubin smoothing is one of the oldest but fast and effective.

If photoscan has a quality metric per vertex on the mesh, it would be interesting to weight the smoothing according to vertex reconstruction confidence.

[Infinite]

+1

Surface-preserving smoothing would be great addition to photoscan. Meshlab has a good library of smooth filters to look at (laplacian surface preserve, hc, two step, taubin). Taubin smoothing is one of the oldest but fast and effective.

If photoscan has a quality metric per vertex on the mesh, it would be interesting to weight the smoothing according to vertex reconstruction confidence.

VERY interesting idea. At the moment I just use the Smoothing and Polish algorithms in ZBrush but I'm sure there must be better solutions. Things that can preserve some data instead of smoothing and destroying all.

[Bendytoons]

+1
I am currently playing with both meshlabs noise reduction as well as a smoothing brush in my voxel editor, but it would be awesome to set this up in PS instead, and the vertex confidence idea is really interesting.

[ajg-cal]

+1

I like how video editing programs allow the use of plugins/addon filters (thinking virtualdub)

a few rudimentary noise filters in which you could specify smoothing parameters would be great

[Matt]

I have used surface extraction filters within LASTOOLS on photoscan data and it works pretty well. Would be good to have something integrated though.

[Wishgranter]

http://graphics.stanford.edu/courses/cs468-10-fall/LectureSlides/08_Simplification.pdf

[Infinite]

http://graphics.stanford.edu/courses/cs468-10-fall/LectureSlides/08_Simplification.pdf

Photoscan already does decimation. What we need is a built in noise reduction tool. Something that can retain some high and low frequency details. That differs from the methods used in Mudbox and ZBrush.

[Wishgranter]

Yes, i know, but this link was becasue of the RESAMPLING - quality and process of the data.... The one here is FAST can say perfect but its not "ideal" - probably will use remesh instead - TRIAS to QUADS....

[jeffreyianwilson]

+1

Surface-preserving smoothing would be great addition to photoscan. Meshlab has a good library of smooth filters to look at (laplacian surface preserve, hc, two step, taubin). Taubin smoothing is one of the oldest but fast and effective.

If photoscan has a quality metric per vertex on the mesh, it would be interesting to weight the smoothing according to vertex reconstruction confidence.

VERY interesting idea. At the moment I just use the Smoothing and Polish algorithms in ZBrush but I'm sure there must be better solutions. Things that can preserve some data instead of smoothing and destroying all.

I use Geomagic heavily for noise reduction in LIDAR scans but I cannot rationalize an additional $4K+ purchase at this time.

Jeff