Scanning a ceramic bust with limited success

[ArtKru]

I've been working on my first PhotoScan project. It is of a porcelain bust of a woman. I've made good progress but the quality isn't what I'd hoped for. I'm hoping someone can give me suggestions on where to go from here. 

I've uploaded one of my photos and a snapshot of the result from PhotoScan. As you'll see, the scan gets increasingly coarse in facial areas that are very even. There are also these weird artifacts on the tip of the nose and chin.

The bust stands about 18" tall. I'm really only interested in the neck and face so I'm not worried about the lower areas that are missing. I've taken 112 photos in 7 circles of 16 photos each.  I'm shooting TIF images with the camera (a Canon 5DMkII) on a tripod using a ring light. (I also tried using even cloudy day room light with long exposures but the results were the same.) The quality of the photos is quite good. The background is black velvet. The bust is on a turntable I've marked with 16 evenly spaced segments (so 22.5 degrees each). I manually turn the turntable so I don't hit the marks with perfect accuracy each time. Probably plus/minus 3 degrees.

In PhotoScan I've left the Preferences at what I think are the defaults. Under General I have

General
Language:English
Default View: Model
Mode: Anaglyph
Parallax:1

OpenCL uses my GeForce 1070.

I'm importing the images and then matching points at the "High" setting with "Pair Presentation" disabled and "Adaptive camera model fitting" checked.  When building the dense cloud I also set quality to "High."  When creating the mesh I left things at Surface type: Arbitrary, Source data:Dense Cloud, and Face count: Medium.

I'd very much appreciate suggestions on where to go from here.

Thanks

[chrisd]

Hi ArtKru,

I believe part of your issue may be due to shiny (reflective) surfaces. A diffused light source might help. If possible, apply some type of dulling spray, though of course the situation may not permit this.

Also, a polarizing filter could help to minimize certain reflections.

[ArtKru]

I couldn't use a matt spray unless I could easily wash it off. That would be tricky.  But the polarizing filter is a great idea. Thanks for the tip.

Art

[Alexey Pasumansky]

Hello ArtKru,

If you are interested only in the model shape you can also use the projected pattern.

[ArtKru]

Thanks for the tip. Where can I learn more about using projected patterns? I looked in the PhotoScan manual and didn't find a reference. A Google search turned up very little as well.

[badger]

Thanks for the tip. Where can I learn more about using projected patterns? I looked in the PhotoScan manual and didn't find a reference. A Google search turned up very little as well.

It would be very complicated if this is the onlt project that requires this. You could consider dusting the subject with darkish powder, graphite or so, which should be way easier to clean out afterwards.

[FLuca]

@ArtKru
Considering the glossiness of ceramic, you have pretty good results.
You will never have perfect mesh out of scanning process, you have to finish it manually.
From my humble experience, scanning and calculating a model is only 20-25% of the time required to generate a finished model.
After photoscan you have to crop the model, remove unwated mesh, close the holes, then retopologise, re-sculpt, remove the noise and correct the texture(s).

[MeSelfie]

Every one want to follow the Precautions to take when the projecting patterns the scan body for good result :

• Number of projected patterns: determines whether the method is valid or not for
measuring moving objects.

• Number of cameras: the method uses stereovision (2 or more cameras) coupled
to a non calibrated pattern used only to get texture on the surface pattern, or a
unique camera coupled to a calibrated projector.

• Axis codification: the pattern is coded along one or two axes.

• Pixel depth: refers to the color and luminance level of the projected pattern (B,
G and C stands for Binary, Grayscale and Color respectively).

• Coding strategy: refers to the periodicity of the set of patterns projected on the
surface (A stands for Absolute and P stands for Periodic).

• Sub-pixel accuracy: determines whether the features are found considering subpixel
precision, thus providing better reconstruction results

• Color : determines whether the technique can cope with colored objects

@For more info contact us
mephoto.in/meselfie-3d-studio