[Leica] Was: R10 news, now: the IR filter

[hlritter at bex.net (Howard Ritter)]

On Nov 2, 2008, at 1:02 PM, Frank Filippone wrote:

> I understand this argument..... the thickness of the IR coating is  
> in the
> order of angstroms....  and at an angle, would be "different" in  
> passband
> than at 90 degrees.... so the images at the corners would have some  
> slight
> change in passband.......IOW, the colors on the periphery would be
> different than in the center.  But it is fixed, and based purely on  
> the
> angle of incidence.  Easily corrected, as the 6 bit coding does for  
> Leica
> now......

The only "color" that would be different at the periphery is IR; the  
IR-cut interference filter has no effect on other wavelengths.  
Therefore, an algorithm that subtracts red (the sensor's  
interpretation of IR) in a graduated fashion from the center to the  
periphery thanks to lens coding would work correctly only if there is  
a predictable component of IR everywhere in the image. It will get it  
all wrong if there is no IR, or nonuniform IR, in the image. Which is  
to say EVERY natural image! The algorithm you're thinking of may be  
the one that corrects for fall-off of illumination at the periphery.  
Since this is a geometric property of the lens, and not color- 
dependent, it is indeed fixed, predictable, and removable. Ditto for  
chromatic aberration. Ditto even for lenses whose glass has a color  
cast. But not for IR.

> So why the fuss over putting the IR coating on and correcting it  
> using the
> coding and algorithms in the camera?.
>
> And why I still do not accept the story........     still looking to  
> be
> convinced......

But, even aside from the effect of the angle of light on the effect of  
an IR filter, the IR color cast CAN'T be corrected by an algorithm in  
firmware or software, because the contribution of IR to an image is  
not constant...not from image to image nor from one part of an image  
to another part...filter or no filter.

Consider a piece of the proverbial black polyester, which reflects IR,  
which registers in the red-sensitive photosites and causes the sensor  
to see red, because it can't distinguish red from IR. This produces a  
reddish color cast. Beside the piece of polyester is a piece of fabric  
whose color to the human eye really is a red-tinted black, and which  
reflects no IR. The two pieces (black + IR, black + red) look  
identical in color to the sensor, but not to the eye. A software  
algorithm that subtracts red from a  predominantly black region of the  
image will do the same to both the color-casted fabric and the true- 
color fabric. Since the firmware has no way of knowing how much of the  
signal from a red photosite is due to invisible IR and how much to  
visible red, the result is an image that shows the IR-reflecting  
fabric in its true color (maybe) and the true-color fabric with a  
green cast!

The only way of removing the IR signal is to remove the IR from the  
light that reaches the sensor. Why would camera manufacturers even go  
to the trouble and expense of putting an IR-cut filter in their  
cameras if the problem were solvable in firmware?

--howard