Could this have been a viable design approach for the M8 sensor?

[runkel]

Since a standard Bayer filter has rows of red and green filters alternating with

rows of blue and green filters, more photosites on a sensor are capturing green

spectrum than seems strictly necessary (although I gather there are sound

reasons for doubling up on green).

 

If half of the green filters were replaced with filters blocking all visible

light and UV the sensor would provide a channel of IR-only data. Infrared would

then be a known quantity, and firmware could then subtract out the influence of

infrared on the RGB photosites. Perhaps this scheme would even allow Leica to

omit the technically vexing IR filter. (This assumes that the filter for non-IR

spectrum could be produced with a normal dyeing process and would not be

susceptible to the same angle-of-incidence problems associated with dichroic

filters.)

 

Might this have been a workable approach? I would assume this is something that

Leica and Kodak would have considered but rejected for sound technical reasons.

Perhaps without the "extra" green photosites it would not be possible to get a

quality 10MP image from the sensor. However, if the M8 design process heavily

favored off-the-shelf components, an approach requiring a new Bayer filter might

not have been explored. If the approach would work in theory, could the M8's

existing sensor be retrofitted with a modified Bayer filter?

[Rob F.]

Matthew, it sounds like you are light-years ahead of some of us--perhaps most of us--on these technicalities. I'm sure I don't know the answer, but I do have a question. Why would the approach you suggest be better than just putting an effective IR filter over the sensor? OK, you said it was "technically vexing." But would not the approach you suggest also be vexing? Would it tie up some of the sensors for IR-sensing purposes? Maybe that's why you said there might not be 10MP available anymore.

 

I didn't know there was such a thing as a Bayer filter. Is that part of a normal digital camera?

[duane_kucheran]

Not a bad idea. My background is electronics with experience in microelectronics packaging and I'll say that it almost never makes economic sense to take any sort of package apart to be reused; it's just easier and cheaper to start fresh.

 

The one downside I can think of is that while the intelligence (signals) are correlated and can be subtracted, noise is random and unfortunately additive so the low light situation would be compromised (~41% more) but not a whole lot. Shannon et al had a whole bunch to say about this. I suppose this step could be deferred to the photoshop phase though. Nevertheless, engineering is a series of compromises that can be made to actually work very well.

 

Cheers,

[runkel]

Rob,

 

<a href="http://www.bythom.com/ccds.htm">This</a> has a good description of how the Bayer filter fits in.<p>

 

My understanding of the IR filter problem is based on reading a lot of internet posts, so I might not have this all quite right and someone will need to correct me. The IR filter is of the "dichroic" type that works by interfering with targeted wavelengths of light. But it is an immutable feature of these filters that they cut out different wavelengths depending on the angle of incidence of the incoming light. Because a rangefinder lens is positioned close to the sensor, light enters the sensor at steeper angles than in a DSLR, especially at the edges of the frame. Because of the angles of incidence involved, a "regular" IR filter would cut out an unacceptable amount of visible light (presumably at the red end of the spectrum). Leica's solution was to use a thin IR filter that leaves the visible spectrum intact, but the trade-off was allowing what has proven to be an excessive amount of infrared through to throw off the sensor.

 

It does not seem that the Bayer filter is sensitive to angle of incidence, I assume either because it lies below the angle-adjusting microlenses, or uses conventional dye-based filter elements instead of dichroic ones, or both.<p>

 

Duane, interesting point about noise. I'm sure there's no way around noise on a fundamental level. Maybe because the IR photosite information would be used only to help the camera know how to "dial down" the readings coming from the RGB sites, there could be clever ways to keep the noise effects from being strictly additive. But that's just speculation on my part in areas way beyond my layman's depth.

[piotr_panne]

I thought the Bayer filter helped block hangovers but nothing can stop the normal dyeing process.

[lutz]

Matthew, the thought had occured to me, too, and I think your idea is a very valid one. But we're talking re-designing sensors and firmware here. Leica might as well buy into a proven and working solution as developed by other manufacturers at this point, don't you think? Since the others must already be doing something along the lines you described...?

[joseph_wisniewski]

Human vision is basically trichromic (red, green, and blue). There's no way to uniformly distribute equal numbers of three colors in a square matrix, without resorting to outrageously complicated patterns. Dr. Bryce Bayer came up with the idea of boosting resolution by considering one of the three colors as "luminance" and doubling its number relative to the other two colors. Since green is the center of the visible spectrum for us humans, it "won".

 

Eliminating half the green as you suggest would decrease resolution 31%. The 10.2mp M8 would have resolution equal to a 5.1mp Bayer style camera.

 

Personally, I think Leica would have been better off with the stronger dichroic filter on the sensor and using software to correct the cyan corners.

[ken_ford1]

Piotr Panne , nov 17, 2006; 08:25 p.m.: "I thought the Bayer filter helped block hangovers but nothing can stop the normal dyeing process."

 

Flag on the play! Double pun - penalty ten yards!