Saturday, January 5, 2008

Extracting useful Information out of Multispectral Images: an attempt

I thought about how multispectral imaging could lead to the enhanced extraction of useful information aside from the fact that unusual, strange or even weird colored images might be produced.

Let's for the moment concentrate on the extraction of information here and the way to present that in a way to make it clearly visible, even to the unskilled reader.

Since it is still cold and dark here, I used a Phalaenopsis orchid flower as the target.

In all cases the UV Rodagon 60mm served as taking lens, focus was not adjusted between shots.

[the following example is intended as a quick demonstration and far from perfect in terms of alignment due to movement, noise etc. Further, I'm not a biologist, so my wording is not correct to denote certain plant parts; please bare with me.]

[as usual a click on an image yields a large version]

1) Visual shot using Baader UV/IR cut filter @ISO200 f16:

2) UV shot using Baader U-Filter @ISO400 f5.6 showing strong UV pattern around the pollen area (round tip in the middle), the zebra pattern mound and the front tentacles i.e. strong UV reflection [directly from the camera, no modifications]:

3) IR shot using B+W 092 filter @ISO400 f5.6 showing even distribution of IR reflection with just a fine zebra pattern i.e. IR reflection [directly from the camera, no modifications]:

4) Combination of 1+2 in such a way as to make the strong UV reflection pattern clearly visible in this case shown as blue / turqouise using my differential mapping technique i.e. UV is mapped into the visible space as blue:

5) Combination of 3+4 in such as way as to also include the soft IR reflection patterns and still trying to preserve the info of no. 4 using the same mapping technique but here IR is mapped in as red:

6) This is for comparison the mapping UV-->B, VIS-->G, IR-->R which in my opinion needs a very skilled reader to see these patterns due to additive mixing of false colors (or one could even say it suppresses it to the unskilled reader):

So what do we have here now in the final result:
we clearly see that the flower exhibits a strong UV pattern around the pollen area (which may serve to attract pollinators), the zebra pattern around the "mound" and thse tentacle like parts which otherwise would have not been visible in the white light shot.

The IR contribution to the overall result in this case is quite even and except some soft zebra pattern does not contribute much, so no. 4 could have even served as a final result - in that case.

This case was meant to express my opinion, that if a meaningful result for multispectral imaging in terms of extraction of information is wanted, the method of generating the final result should be chosen accordingly and the way of combining the different spectral images should be governed by that - just my 2 cents.

Stay tuned, more will follow on that fascinating subject...

More info on this very interesting field may be found on my site