M8 Cure - test my thinking on this

[pico]

Advertisement (gone after registration)

The essence of the IR problem is that the camera cannot have knowledge of the objects it 'sees'. For example, it cannot tell that there is a synthetic fabric item that is super-reflective to IR on the sensor. Presuming a front lens filter is not acceptable to users, what of the following approach?

 

What if the camera first measured the nominal normal exposure, then emitted a burst of IR toward the subject and compared the two readings. A highly IR reflective object would create an unusual luminance deviance from the norm. Two or three things could then be done: 1) issue a warning in the finder 2) apply local adjustments to the affected pixels, or with a configuration setting, do nothing (and in any case, record it to EXIF.)

 

I doubt there's room in the M8 body for the device, but there would be room in a flash, and an opportunity there because flash transmits a lot of IR and might be the worst case. And the device could be tagged to the top or bottom of the camera, ala old leica gizmos.

 

I am trialing the concept of the comparitive readings as a thought experiment. Might it work?

[Vivek Iyer]

I am trialing the concept of the comparitive readings as a thought experiment. Might it work?

 

If it has not worked for you, it probably will not work on a real camera, Pico.

 

Far out.

[pico]

If it has not worked for you, it probably will not work on a real camera, Pico.

 

Far out.

 

It works for me as a thought experiment, and that's why I'm asking for expert critique, hoping that experts chime in.

[Vivek Iyer]

I am afraid all the "experts" are elsewhere on the 'feisty' site (you know where).

[Guest rubidium]

No, in principle this wouldn't really work. Getting mathematical for a moment, you are suggesting using two measurements to determine three unknowns, which is an "underdetermined" problem. Considering just a single "red" pixel in the Bayer matrix, the unknowns are: A, the intensity of ambient light in the visible red portion of the spectrum that we wish to keep; B, the intensity of ambient light in the IR region that we wish to reject; and C, the intensity of IR light generated by your supplementary source. The first frame that you record captures intensity U=A+B for that pixel, and the second frame that you record (assuming nothing has moved) captures intensity V=A+B+C for that pixel. By differencing, (V-U) can help you to remove the effects of C from the second image, but there's no way to remove B, which is the overall objective of the effort.

 

Really there is only one way to "unravel" a mixture of desireable light in the red region of the visible spectrum from undesireable light in the near IR region that simultaneously impinges on a sensor that is sensitive to both - without performing filtering ahead of the sensor plane. Unfortunately, the state-of-the-art in sensor technology is many orders of magnitude short of being able to support this. To use the sensor itself to separate the two, without a priori filtering, would require the sensor to have an electrical bandwidth comparable to the frequency of light (e.g. 4x10^14 Hz at 700 nanometers), and to Nyquist sample the incident energy for some fraction of the exposure. Then digital signal processing techniques could be applied to effectively remove the unwanted spectral components. Today's CCD's and similar semiconductor sensors like the Kodak KAF-series have bandwidths on the order of 45MHz - seven orders of magnitude short of what would be needed to do the filtering "digitally."

 

Jim