The quest for detail

Hands-on experience with the Minolta VIVID9i 3D laser scanner

The desire for more detail has been perfectly described by Paul James on his article about microscopy, I quote: "Subconsciously there will always be a desire to get closer to the subject". This is certainly true for me.
The work I present here was carried out during the practical courses of the International Summer School; Computer Vision for Cultural Heritage 2007 at the Pattern Recognition and Image Processing Group of the Vienna University of Technology. To see how the Minolta scanner would perform on scanning small objects I used a flint arrowhead and a clay pipe fragment. The data was acquired during several hours of scanning, without any prior experience with this equipment and software. Most likely, an experienced operator could get better results. Afterwards, I processed the data using Geomagic Studio 9 which took much more time.

Lessons learned from this exercise:

• For the small objects (< 10 cm) I used, increasing the resolution (and accuracy) by an order of magnitude would be needed to capture most of the details archaeologists are interested in. Something like 10 micrometer resolution might be enough; possibly other scanners are capable of this.
• A 'realistic' texture would probably improve the visual appearance and 'hide' some of the loss in geometric detail. To acquire useful textures you would need much better control over lighting conditions and create a little photo studio around the object.
• It is difficult and sometimes impossible to scan into holes, also scanning thin ridges and thin objects is difficult.
• The same process that filters out the noise during the merging of scans, also removes important details. Maybe the merging can be improved?
• For accurate measurements on the scans I would test or calibrate the equipment first, because with both objects the scan had smaller dimensions than the real object.

The flint arrowhead

This arrowhead is a modern production by an experienced flintknapper. A real 'ancient' arrowhead would most likely have a shinier surface due to degradation processes. This is still an interesting object to scan because the material is semi-transparent which might pose problems for the scanning when the laser beam penetrates the surface, especially because it is also thin. Details that would be nice to capture are the 'flake scar patterns' visible by the ridges where the scars meet. Other things would be the scar ripples and percussion bulbs, but this is not expected with this small object and the resolution of the Minolta.


First scan sequence

• Used Tele-lens at a distance of about 0.6 m.
• Positioned fragment using 'Pritt poster buddies'.
• 60 degrees per scan, with the setting to remove 5 deg. from border

After cleaning I registered globally. The each side seems to be registered, but the two are overlapping on both the points' sharp edges. Because it is so thin, there is no scan that has triangles of both sides. Or maybe it is also because of the transparency, both side are scanned sub-surface and therefore overlapping? To be sure, I should acquire an extra scan with a thinly sprayed surface or even place some other object on the table to register...but I didn't.

registered globally, see blue overlap of opposite side

other side

Merging them now, gives bad edges, and not registering before merging does not really help. The next thing I tried was to register both sides separately and then combine them manually. The result looks better, but is it a true representation of the object?

The overlap of the two sides.

Moved one side 0.8mm along z axis and 0.5mm over x-axis.

Instead, I tried another way of scanning. Using an angle of 90 deg. so only two scans per side, each with about 45 degrees inclination angle with surface. And of course one scan sequence for the base and another for the tip. Registered both side separately and then combine them. The result looks better, but did not have the detail I would like to see.

scan 4	- scan 5
final result after cleaning and hole filling	Reduced to 50% = 6020 triangles and exported to vrml2 (333kb). Note: To view the VRML model, you can use Cortona VRML plugin

Measurements on the resulting model:
Distance tip-base = 20.6 mm, volume = 337 cubic mm.
Measured on real object (Scala INOX steel caliper):
Distance tip-base = 22.0mm (0.5) => at least 1mm shorter when scanned.

The clay-pipe fragment

This is a small ceramic object. The shininess is the result of the surface treatment (burnishing). It also contains some other interesting details; the heel mark (number 33) and mark on side of heel.



Two scan sequences were acquired.
First scan

The unregistered scans of first sequence.	One of the scans contains detail: number 33 on the heel.
Global registration works fine.
After merging the detail is lost, and a hole at bottom is also visible.	The rim is badly scanned.

Second scan

Unregistered scans of second sequence, whole part resting on the table is badly scanned.	Some detail ('shield with S') on side of heel is visible in a single scan.
Registration is fine.	But details lost again.

In order to keep the details but lose the noise, I tried to merge the details (after registration) without filtering. This did not result in an acceptable model.

The final result.

Result without trying to keep details.

Reduced to 9478 triangles and exported to vrml2 (533kb) . Note: To view the VRML model, you can use Cortona VRML plugin

Measurements on the resulting model:
diameter = 20.4, but it is difficult to measure that accurate by placing points in 3D. length = 47.3
Measurements on the real object
outer rim (opening) diameter = 21.0mm (0.5) length (from rim to base) 49.0 mm