New filter combo for digital Aerochrome with Midopt triple bandpass

[Fedia]

The following experiments are inspired by the findings of @Christoph

 

Here is a series of pictures with my full spectrum canon 1200D and a stack of these three filters :

 

• Midopt Triple bandpass 550/660/850nm
• Lee "Flesh pink" gel
• GRB3 from Tangsinuo

 

The transmission curve of the three filters combined should look something like this :

 

 

Both the green and the IR spike are heavily filtered. When the TB is used alone, it seems to suffer from a very weak red transmission : without the GRB3 the red channel records as much (if not more) IR than red, requiring IR subtraction of 100%. The green spike transmission to the contrary is very powerful. To the eye (at least mine) the filter is green. I'm not sure how bright humans are able to percieve red at 660nm, its pretty deep already.

 

I then decided to reduce the transmission of IR and green equally to give room for the red transmission. The GRB3 is used to minimise IR contamination in the RED channel and the Lee gel is used minimise green contamination in the blue/IR channel and to make it more or less match the level of output of the red channel. The blue channel stays very underexposed compared to the two others, and is brought back in white balance (pretty extreme : lower than 1900Kelvin).

 

The exposure value with these filters ranges from 1/50s to 1/100s in sunlight at f5.6, 200iso. It's a very dark combo.

 

All the pictures are channel swapped in Resolve from the camera Jpegs, the process is video ready. No saturation was added and zero IR substraction is needed.

 

 

 

 

 

 

 

 

 

 

 

 

The fact that IR transmission is cut by 90% by the GRB3 causes some green leaks to be percieved in the Blue/IR channel. This leads to the need to apply a hue correction to the sky in order to make it look properly blue and not purple-ish. This process is simple and non-destructive.

 

The original color of the sky after channel swap :

 

 

 

 

Important notice :

 

The channel swap causes a significant decrease in contrast and micro-contrast. For a moment I thought this was due to the Jpeg compression but it is not.

 

Do you remember this law ?

 

Y = 0.1 Blue + 0.6 Green + 0.3 Red

 

It is actually crucial in understanding why the images look better unswapped. It describes how the human eye is sensitive to luminosity. The green value is far more decisive in the percieved brightness of an object than the red and blue values.

 

 

A quick exemple

 

Unswapped image :

 

 

Swapped image : it looks less detailed and dynamic. The contrast beetween the bright grass and the dark trees is lessened

 

 

Swapped image but with the "preserve luminance" box ticked : the image is sharper and more alive. (Click on the image and use the viewer to compare both instantaneously.)

 

 

Explanation :

 

The grass in the unswapped image appears turquoise. Let's make it cyan for the sake of clarity. The trees appear blue.

 

Cyan RGB values are (0; 255; 255)

Blue RGB values are (0; 0; 255)

 

The minimum luminance(Y) is 0 and the maximum luminance is 1.

 

Y(cyan grass) = 0.1x Blue(255/255) + 0.6x Green(255/255) + 0.3x Red(0/255) = 0.7

Y(blue trees) = 0.1 x B(255/255) + 0.6 x G(0/255) + 0.3 x R(0/255) = 0.1

 

in the unswapped image the contrast beetween the brightness of the trees and the grass is important : Y=0.1 versus Y=0.7.

Now, let's do the same for the swapped image where the grass is magenta and the trees are red.

 

Y(magenta grass) = 0.1 x B(255/255) + 0.6 x G(0/255) + 0.3 x R(255/255) = 0.4

Y(red trees) = 0.1 x B(0/255) + 0.6 x G(0/255) 0.3 x R (255/255) = 0.3

 

in the swapped image the contrast beetween the brightness of the trees and grass is less important than in the original image (Y=0.3 versus Y=0.4)

 

 

In this chart the colors are ranked by luminance. white Y=1, yellow Y=0.9, cyan Y=0.7, green Y=0.6, magenta Y=0.4, red Y=0.3, blue Y=0.1, black Y=0

As you can see here the brightness of Magenta and Red is very close as opposed to cyan and blue that are very far apart.

 

So that's why channel swapping sometimes makes the image lose quality.

 

A solution to this is to tick the box "preserve luminance" in the channel mixer. It doesn't work in every situation since it makes the yellow objects turn way darker, leading to an unnatural look. Ticking the box also makes the sky way brighter. It's a tradeoff that has to be made for each individual picture.

 

In the selection I posted above, a few have the box ticked and most don't.

 

 

[colinbm]

Impressive work Fedia.
I will need to re-read this to get it into my tired old brain.

[Fedia]

Thanks Colin !

 

 

[Fedia]

By the way, can anyone tell me if other software has this "preserve luminance" function in their channel mixer ? I don't know a lot about photo software.

 

[Andy Perrin]

These look fantastic in terms of color balance, but the overall brightness of the images is VERY low, at least on my monitor. Most things are buried in shadow. I would use a bit of a gamma curve after all the other adjustments are finished on it.

[lonesome_dave]

Very impressive Fedia.

 

Makes me appreciate how easy I had it back in the film days when the only thing I had to deal with was which filter to use with Ektachrome IR--- the #12, #15, #21 or #25.

 

Looking forward to eventually having a fairly straighforward digital Aerochrome solution that can be applied to most cameras. (If that's even possible).

 

Appreciate all your efforts on this.

 

[Fedia]

44 minutes ago, Andy Perrin said:

These look fantastic in terms of color balance, but the overall brightness of the images is VERY low, at least on my monitor. Most things are buried in shadow. I would use a bit of a gamma curve after all the other adjustments are finished on it.

Yes, I noticed it too. My images always turn out too dark when I edit in Resolve. The software doesn't invite you to make your image pop like Lightroom does for exemple.

[Fedia]

18 minutes ago, lonesome_dave said:

Makes me appreciate how easy I had it back in the film days when the only thing I had to deal with was which filter to use with Ektachrome IR--- the #12, #15, #21 or #25.

Thanks Dave ! That must have been amazing times !

[Kai]

A very profound work and reflection! Very impressive.
Thanks for sharing Fedia ;)

[Fedia]

To verify what I was I was saying earlier about the effect of channel mixing on contrast, I drew a face on paint with a cyan line (0; 255; 255) on a blue background (0; 0; 255) :

 

 

Then applied the digital aerochrome channel swap to it :

 

 

Same thing with the preserve luminance box ticked :

 

 

Now the same three image in the same order but converted in black and white :

 

Background Y= 0.1, Face Y=0.7  → High contrast

 

Backround Y=0.3, Face Y=0.4  → Low contrast

 

Preserve luminance box ticked. The image should look the same as the original but obviously doesn't. The contrast is nice though.

 

It is interesting to see that the software doesn't dare to really apply the luminance of blue (Y=0.1) to the red background because if it did, it would be so dark that we probably wouldn't identify the color as red. The preserve luminance function has limitations but performs well at recovering contrast even if it has to cheat to preserve the colors.

 

[Fedia]

The effect of channel swapping on noise.

 

Doing digital aerochrome implies using a deep yellow/orange filter and sometimes other fancy combinations like I did for the pictures above. In all cases the filters need to cut all the blue wavelength out in order to record IR in the blue channel. This always leaves the blue channel underexposed compared to the other two making this channel also the noisier.

 

In the unswapped image the noisy blue channel has a luminance weight of 0.1 in the overall image. But as soon as its information is transfered to the Red channel when the channel swap is applied that noisy channel all of a sudden has a luminance weight of 0.3 in the overall image. That's like multiplying its luminance by x2.5. It is no wonder that the noise becomes visible after the channel swap.

[Andrea B.]

Excellent analysis, Fedia. Thank you for this.

 

If a photo software does not have a preserve luminance capability, what can we do after swapping channels to improve the appearance?