Do you replace the IR-filter with another filter in front of the sensor or just put the right IR filter on the lens?
Hi Tomas,
At the moment I don't use either; in the example I attached above I let the full visible + IR spectrum through. This is best for photon-starved astrophotography.
But I will probably soon pick up a lens-mounted filter, to play around with real deep-IR landscape stuff. Above 850nm, the RGB CFA filters are all equally transparent, so I should be able to do true B&W photography without even needing to do any Bayer interpolation in RAW processing (rather like the PhaseOne P45+ Achromatic back).
Ideally, one would have an IR filter cut to fit over the sensor, instead of over the lens, with the advantages of
- you can see what you're shooting
- no filter changes with lens changes.
This would be my preferred solution for daytime IR, and it's something I'll aim at doing in the longer term.
One might also think that, with the right refractive index and thickness, a new filter over the sensor would preserve the original optical path length, so that AF would still be accurate.
But since the AF is sensed in visible light while the image is formed in IR light, AF would not be accurate, unless the lens is apochromatic right down to the IR - as Jack has already alluded to above. Jack also makes a very good point about how cameras which use their imaging sensor for their AF, like the micro 4/3rds, inherently avoid this problem and reach their best IR focus.
Nor does one necessarily need a filter over the sensor to reach infinity with the lenses; you know that little red mark for IR infinity that every lens has? It moves the lens forward from the film/sensor. If instead one removes the filter from the front of the sensor, one achieves much the same change to optical path length, so that one then uses the normal visible-light infinity mark for IR.
The corollary is that in a completely unfiltered system, if the IR component is satisfactorily focused, the focus for the visible light component (actually a small fraction of the total light transmitted) is not optimal. I am of course aware of this softening of net focus.
However this doesn't concern me unduly because it is important to be aware of yet another effect which softens IR images - an irreducible effect which cannot even be fixed with the right filters, optical path lengths and uber-hyper-APO (or all-reflecting) lenses! It is the "Diffusion MTF issue", explained here - http://bit.ly/9mCrvb . Basically, IR photons penetrate so deep into sensors designed for the visible spectrum that the photoelectrons they release wander (diffuse) all over the place and end up in the wrong pixels.
You know the way that, up to a point, things scale in imaging? - i.e. smaller sensors have fewer mm but the lenses designed for them tend to resolve more lp/mm, so quality scales reasonably well? Well, Diffusion MTF breaks this scaling in the IR. The degree of electron diffusion is more like a non-scalable absolute; in fact, if anything, it's a stronger and worse effect in small-pixel sensors. This is one of the main reasons why a larger (MF) sensor is best for IR photography.
Ray