In my neverending quest for realism in photographic imaging, I've been researching High Dynamic Range (HDR) imaging for years. In that quest I only recently (I have too much to read, too little time) ran into this document:
http://www1.cs.columbia.edu/CAVE/tomoo/RRHomePage/download/mitsunaga_nayar_cvpr99.pdfWith some out-of-the-box lateral thinking, I wondered if that procedure could be useful in determining the OECF. An accurate OECF is required for various ISO standard tests, including SFR and noise/dynamic range determination.
To me it seems that with e.g. three exposures (one normal, one 1 stop under, and one 1 stop over) of a stationary high dynamic range scene (e.g. a room interior with a window looking out over a sun lit outdoors plus sky, or a side lit stationary 3D object of average semi-specular reflection).
Obviously one would attempt to use a subject that, with contrasty lighting, would exceed the dynamic range of the recording medium, but then that can also be achieved by taking more exposures with more extreme under/over-exposure until clipping occurs.
The only real restriction is that the two (or more) exposures need to be in exact registration (hence the stationary subject), but even that could be achieved/verified with e.g. this sub-pixel accurate procedure:
http://www.anyhere.com/gward/papers/jgtpap2.pdf (amazingly simple once you grasp the underlying issue).
The beauty of OECF determination with multiple exposures is that there is no real need for expensive calibrated targets and critical uniform lighting of that target. Even limited dynamic range targets like a Q13 can produce full dynamic (media) range results!
Also issues like lens vignetting, or in-exact shutter behavior, or non-linear tone-mapping, or ISO settings, become a moot point, because the procedure is self calibrating. Even the accuracy can be influenced by changing the exposure ratio difference(s).
Any thoughts?
