Re: 3d Reconstruction from Stack

Posted by Philip Ershler on Jul 06, 2011; 5:41pm
URL: http://imagej.273.s1.nabble.com/3d-Reconstruction-from-Stack-tp3684002p3684005.html

On Jul 6, 2011, at 11:09 AM, Johannes Schindelin wrote:

> Hi Andrew,
>
> On Thu, 23 Jun 2011, Andrew E. Meyer wrote:
>
>> We have an application where we are taking a stack of pictures of a fuel
>> injector seat and spray orifices, and trying to reconstruct a
>> 3-Dimensional image for use in verifying that the hardware was made to
>> spec.  We are using a microscope with a short focal length, so that the
>> features in the stack appear to grow in size as we get closer to the
>> object.  (think of a star field animation, where the nearer stars go
>> racing to the edge of the picture as the spaceship goes through the star
>> field).
>>
>> First, is there a name for this "star field" effect by which I could
>> search for information?
>>
>> We have tried using stack focuser and EDF to create a three dimensional
>> map from our stack with very limited success-the images are very noisy
>> and don't correspond to the reality which we can measure, so we cannot
>> trust them for the stuff we cannot measure.  Maybe I'm just not using
>> these extensions properly?  I'm new to this, but it is my understanding
>> that these techniques determine position of each point based on how
>> sharp the focus for the point is, so moving features might be a problem
>> for them.
>
> I imagine the problem is that your stack slices are not really in and out
> of focus.
>
> So maybe you need something completely different: you could use SIFT
> features in every slice to generate robust landmarks, and then use a
> RANSAC to find the most likely 3D locations of the landmarks.
>
> The last step would be an inverse mapping from the recorded stack into a
> 3D height-field.
>

This might be a totally "off the wall" idea, but I'm wondering if a simple cofocal imaging system might be an alternative way to image the spray pattern. If a fluorescent dye were to be used, the pattern could easily be sectioned into slices of predictable thickness. I'm not sure whether correction for point spread issues (a common problem in confocal microscopy) would have to be carried out or not.

Just a thought,

Phil