Responses of Midwest Optical PB Series NIR Filters for Tricolour Process

[Tom Lewis]

I'm looking at possible filters to use for tricolour images  Likely many of you have already done this or perhaps could simply visualize without needing the graphic.  But I'm sharing here the responses for the Midwest Optical PB Series NIR Filters because I think it's nice to see them all on one chart.  I made the curves solid color for the ones I'm more interested, and dotted for the ones I'm less interested in.

 

Theoretically, the combination of the BP735, BP810, and PB880 look good to me since they provide good overlap so that at least two bands register each wavelength in much of the range, which I hope will make intermediate hues well encoded. 

 

I do worry that the amount of content that is within the coverage of the BP735 may overwhelm the image, but of course the gain for that channel can be adjusted.  
 

Midwest Optical PB Series NIR Response_Spreadsheet.xlsx

[Andy Perrin]

Looks good to me. Do you have a monochrome camera you can use, or a deBayered one? The Bayer will certainly affect the images made with first two filters and trying to recombine the RGB channels has never been entirely satisfactory. For one thing, how should each channel be weighted? The traditional “desaturate” command uses an uneven weighting based on human vision, but one could equally well argue for using (R+G+B)/3… The last filter will not be affected by the Bayer since most Bayers become transparent there. Food for thought. 

 

 

[Tom Lewis]

25 minutes ago, Andy Perrin said:

Looks good to me. Do you have a monochrome camera you can use, or a deBayered one? The Bayer will certainly affect the images made with first two filters and trying to recombine the RGB channels has never been entirely satisfactory. For one thing, how should each channel be weighted? The traditional “desaturate” command uses an uneven weighting based on human vision, but one could equally well argue for using (R+G+B)/3… The last filter will not be affected by the Bayer since most Bayers become transparent there. Food for thought. 

 

 

Nope, I don't have an originally monochrome camera, or one converted to monochrome.  In fact, I'm still waiting for my plan-Jane full spectrum conversion.  So, right now my interest is only theoretical.  

 

Is the reason it is difficult to combine the three channels below about 900nm because the CFA filters have such different responses according to wavelength, whereas above 900nm, the three filters have more similar response?  And is the more similar response above 900nm another way of saying transparent as you have mentioned?

[Tom Lewis]

I noticed an alignment error in the chart which resulted in a problem with how the curves were displayed.  So, I fixed that error, and added the PB850.  I also created another chart for MaxMax filters.  The MaxMax curves are based on printing out the curves from their website and then manually interpolating values from the printout.

 

NIR Filter Responses_Spreadsheet.xlsx

[Andy Perrin]

11 hours ago, Tom Lewis said:

Is the reason it is difficult to combine the three channels below about 900nm because the CFA filters have such different responses according to wavelength, whereas above 900nm, the three filters have more similar response?  And is the more similar response above 900nm another way of saying transparent as you have mentioned?

The CFA filters are in addition to your filter, which means what you see in the R, G, and B channels is the product of (as in, you multiply) the transmission curve from the IR filter by the curve from each Bayer channel color. This will give a different effective filter in each channel. The picture you really want is the one with the Bayer not present. But that’s not what you are getting, so you will have to do your best to reverse the process and approximate the unBayered photo. 
 

And is the more similar response above 900nm another way of saying transparent as you have mentioned?
 

It’s a direct consequence of the Bayer filters being mostly transparent above 850nm or so. They all pass the same amount of light once they become transparent. 

[Tom Lewis]

Thanks @Andy Perrin

[Stefano]

As for my experience, I have done some IR TriColour indoors using LEDs (730, 850 and 940 nm), and I have done it in UV, using filters that don't overlap. The LEDs, being fairly monochromatic, effectively don't overlap too.

 

I think that, if you want a more "natural" image, some overlap in the channels is needed. Of course, since you are dealing with false colors, there's no strict rule you have to follow, but if you want to take an image of a spectrum (such as a rainbow), or LEDs of different wavelengths, that's the most natural setup in my opinion.

 

Look at the spectral response of the cones in the human eye, and how much the M (green) and L (red) cones overlap: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://en.m.wikipedia.org/wiki/Cone_cell&ved=2ahUKEwjx_6vvv9aEAxWh8LsIHU0DAQwQFnoECBcQAQ&usg=AOvVaw1D-4g3savBGYxgPtXSSaFe

 

I don't know how much a Bayer filter will affect the image. Even if the RGB channels will respond differently, I don't think it's going to be a big problem.

 

Also, since cameras see up to ~1100 nm, you could widen your spectral range, with a filter centered around 900-950 nm (like the MaxMax filter). A wider spectral distribution will inevitably bring more color separation, as well as probably making water more blue.

[Tom Lewis]

Thanks, @Stefano