Revision as of 17:21, 3 November 2006

Extended depth-of-field image showing part of a 10-pence coin (compare with 1.39 MByte video of focus stack)

Depth from Focus

3D surface metrology

• In principle it should work with any microscope, which has a motorized z-drive and a digital camera (see Mimas video input).
• The results will be even better, if the illumination optics of the microscope can project a pattern.
• Non-destructive measurement of surface profiles
• With our experimental settings we observed (of course the result will depend on the quality of the microscope)
  • Vertical resolution <math>\ge 0.2\ \mu m</math> (depending on aperture-size, magnification, projection-pattern and the surface properties of the object).
  • Lateral resolution <math>\ge 2\ \mu m</math> (depends).
• Open Source (you are free to improve the code yourself if you redistribute it).

In general one can say: The lower the depth of field, the higher the resolution of the reconstruction. With high magnification (assuming constant numerical aperture) the resolution of the reconstruction goes up. The trade-off is that the reconstruction will cover a smaller area. This can be overcome by lateral stitching (e.g. cogwheel below).

Demonstration

Here are some typical microscope images (showing a surface, which has been shaped using a power beam).

First surfi-sculpt object, Leica DM LAM

Second surfi-sculpt object, Leica DM LAM

Piece of Suevit (enamel like material from meteorite impact) from the Nördlinger Ries, Leica DM RXA

Micro-camera image of a hair on top of a laser printout

Surface of 10-pence coin, Leica DM LAM

Using a focus-stack one can compute images with extended depth of focus:

Extended depth of view for first object

Extended depth of view for second object

Extended depth of view for Suevit (fringes have been removed manually)

Extended depth of view for the hair

Extended depth-of-field image showing part of the 10-pence coin (compare with 1.39 MByte video of focus stack)

If the surface can be illuminated properly, one can even do a 3D-reconstruction of the surface:

3D reconstruction of first object (742kB video)

3D reconstruction of second object (725kB video)

[[Image:suevit20.png

180px|3D reconstruction of suevit (1.4MB video)]]

3D reconstruction of hair (1.4MB video)

3D reconstruction of 10-pence coin surface (profile is amplified 5 times)

Red-cyan anaglyph image of first object (2.0MB video)

Extended depth of field image for a piece of sugar (compare with 532 kByte video of focus stack)

Extended depth of field image of glass fibers (compare with 1.09 MByte video of focus stack)

Reconstruction of 0.6 mm cogwheel using multiple focus sets

As the idea for the algorithm was fixed already, it was possible to implement the algorithm as a command-line tool in less than 4 days, using existing Mimas-software (exspecially the operators for boost::multi_array).

As this is a "quick hack", there's still lots of space for improvements.

Download

The software for estimating height-maps and images with extended depth-of-field is available for free (under the LGPL)! You first need to install version 2.0 of the Mimas Real-Time Computer Vision Library to be able to compile and run depthoffocus-0.1 (652 kByte). The software also comes with sample files to generate photo-realistic 3D-reconstruction using POVRay!

Links

• Anaglyph images
• german diploma thesis about depth of focus (also available as PDF)
• Mechrob-paper Focus set based reconstruction of micro-objects
• Depth-from-focus applied in forensic science, MRT Karlsruhe
• Focus stitching software CombineZ5 and examples at http://www.micropics.org.uk/