What a monochrome sensor will show you

[Brian]

The particular CMOS sensor used in the new camera was already available in color and monochrome. I have read (not verified with the manufacturer) that the CMOS sensor used in the full frame camera was available only in color. The CMOS array used in the Nikon D4 and Df was available in monochrome and color, Nikon offered Microscope cameras using the same 16MPixel full-frame sensor. Nikon made a prototype Df-Monochrome, but did not produce it. The Nikon rep told me this, the first day the Df was out- when I bought it.

As stated before- I run most (95%) of M Monochrom images through my own software that adds a gamma curve. Very little processing after.

 

[Flektogon]

View attachment 9579
Fortran Output.

Very nice picture Brian! However, I have further questions. So, this picture was taken with a camera/sensor equipped with the Bayer filter. Now, you are talking about the custom filters, what does it mean, that you really use coloured filters on the lens and then for each picture/filter you apply your own, filter specific demosaication? That's rally a very complicated way to make B&W pictures.
I am surprised that you (anyone) still uses Fortran! Well, I personally hated it from the first moment. At the university we were taught Algol, but I started with the assembly languages, later Pascal but quickly switched to C. Even today (I am retired) I am still playing with the programming, but mainly in C# and Python. Anyway, I see the only reason to use Fortran because of the strong matrix arithmetics. But Python, for example, with the numpy library (which is an integral part of any Python packages), is at least as good as that, by God, long time ago forgotten programming language of yours .

 

[Brian]

In Fortran- I read the DNG file into a 1-D buffer, parse the Tags, find the image dimensions, find the start of the image- then pass that point to the processing subroutine along with the columns and rows and declare it a 2D array. I coded my first digital image processing algorithm in 1980.

I have the spectral response of the KAF-18500 sensor for the M9 and KAF-10500 for the M8. I used the Yellow Y48 filter- means the Blue pixels pick up only green. About 1/2 or so the sensitivity. Then equalize the values- so it basically gives 3 pixels of green and 1 of red for every 2x2 bayer pattern. Some improvements could be added- like a more sophisticated interpolation. I wanted to see how far a color camera could be pushed to produce monochrome. This is after buying the M Monochrom. I called Kodak when the M9 came out and asked for one. The engineer at Kodak was all for it, preaching to the choir. He said they had a meeting with Leica in a month and he would bring it up.

Using a Monochrome camera for monochrome images produces a cleaner file that requires much less post-processing. You can get good results from a modern high-pixel count camera. I had the M9, and then bought the M Monochrom- because I knew in advance what the real differences would be. No Interpolation Artifacts - that alone makes it worth while to me. I spent a lot of time writing image processing code, including what would be called "Machine Vision". Stupid artifacts. Mess up good algorithms. And I just like using the monochrome camera.

Plus, I just like writing code.

Leica Showcase - Leica M Monochrom

1936 Carl Zeiss Jena 5cm F1.5, wide-open with a Zeiss Medium Yellow filter. Straight exports from LR6.

cameraderie.org

The older CCD cameras can pick up bad columns, and have to be sent to Leica for "remapping". That means they just call a column dead and average over it. I fixed it in post, did an algorithm that calculates the DC offset of the bad column and adds it back in. Something similar could be done when converting a color image to monochrome, knowing the spectral response, "guessing" the color based on nearest neighbors, and restoring the offset.

I especially like that Pentax offers uncompressed DNG files. Should be able to adapt my Gamma curve to it.

These are Monochrome conversion that start with a Y48 filter on the camera, custom demosaic. This was a test of an early version of code.

• Leica Camera AG - M8 Digital Camera
• ƒ/5.1
• 1/500 sec
• Center-Weighted Average
• ISO 160

• Leica Camera AG - M8 Digital Camera
• ƒ/4.2
• 1/3000 sec
• Center-Weighted Average
• ISO 160

 

[Dino26323]

In Fortran- I read the DNG file into a 1-D buffer, parse the Tags, find the image dimensions, find the start of the image- then pass that point to the processing subroutine along with the columns and rows and declare it a 2D array. I coded my first digital image processing algorithm in 1980.

I have the spectral response of the KAF-18500 sensor for the M9 and KAF-10500 for the M8. I used the Yellow Y48 filter- means the Blue pixels pick up only green. About 1/2 or so the sensitivity. Then equalize the values- so it basically gives 3 pixels of green and 1 of red for every 2x2 bayer pattern. Some improvements could be added- like a more sophisticated interpolation. I wanted to see how far a color camera could be pushed to produce monochrome. This is after buying the M Monochrom. I called Kodak when the M9 came out and asked for one. The engineer at Kodak was all for it, preaching to the choir. He said they had a meeting with Leica in a month and he would bring it up.

Using a Monochrome camera for monochrome images produces a cleaner file that requires much less post-processing. You can get good results from a modern high-pixel count camera. I had the M9, and then bought the M Monochrom- because I knew in advance what the real differences would be. No Interpolation Artifacts - that alone makes it worth while to me. I spent a lot of time writing image processing code, including what would be called "Machine Vision". Stupid artifacts. Mess up good algorithms. And I just like using the monochrome camera.

Plus, I just like writing code.

Leica Showcase - Leica M Monochrom

1936 Carl Zeiss Jena 5cm F1.5, wide-open with a Zeiss Medium Yellow filter. Straight exports from LR6.

cameraderie.org

The older CCD cameras can pick up bad columns, and have to be sent to Leica for "remapping". That means they just call a column dead and average over it. I fixed it in post, did an algorithm that calculates the DC offset of the bad column and adds it back in. Something similar could be done when converting a color image to monochrome, knowing the spectral response, "guessing" the color based on nearest neighbors, and restoring the offset.

I especially like that Pentax offers uncompressed DNG files. Should be able to adapt my Gamma curve to it.

These are Monochrome conversion that start with a Y48 filter on the camera, custom demosaic. This was a test of an early version of code.
View attachment 9636View attachment 9638View attachment 9635View attachment 9637

This is intense stuff.
Black and White has many many ways of interpretation where it is not written off as completely wrong.

As I have said before. It is many of the test shots from Leica on Dpreview that fascinate me. I am interested in just the luminosity readings without a filter attached to the lens especially on city shots. ( Landscapes likely mostly need strong orange or red filters but it is all guesswork at the moment. )

Some are fascinating. I actually enjoy grain at 100% in black and white so the cleaner high iso is not often that big a deal for me but I guess one could experiment with things rarely captured properly if one had such a camera. The whole concept is fascinating having seen the test shots on my screen.

Regards Dino26323

 

[Brian]

Film and digital sensors have a very different spectral response than the human eye. The eye is not sensitive to blue, film and detectors see well into the UV range. "Way back when" 30 years ago Kodak CCD sensors were not very sensitive in blue. A lot of work went into improving the performance.

To make a monochrome image look more like how it would be seen with the eye, Yellow filters are used. To enhance contrast, and make it "more dramatic", use orange and red.
If you want to catch all the photons- no filters are best. Whether for low-light, or maximum sensitivity, or for measurements. I'm old enough where film was used to convert density on film to "Watts/Steradian". Use film to capture an event, scan it, and convert to a measurement knowing the response curves. 1979. The "Scanning Densitometer" filled a room. By 1982- went to digital sensors for the same type work.

Some old (1930s, 1940s) books on the use of filters are very helpful when using one of these new cameras. Cheap on Ebay. The "WWW" happened well after color photography was the norm.

 

[Brian]

Getting exposure correct with a monochrome camera takes some practice. The Photocell used to set the exposure has a different response from the actual sensor once you get into orange and red.

These are with a Nikon R60 filter, Voigtlander 35/1.2 Nokton, and the Leica M Monochrom v1.

• Leica Camera AG - M Monochrom
• Leica Summilux-M 35mm f/1.4 ASPH.
• 35.0 mm
• 1/1000 sec
• Center-Weighted Average
• Manual exposure
• 1
• ISO 320

• Leica Camera AG - M Monochrom
• Leica Summilux-M 35mm f/1.4 ASPH.
• 35.0 mm
• 1/250 sec
• Center-Weighted Average
• Manual exposure
• 1
• ISO 320

• Leica Camera AG - M Monochrom
• Leica Summilux-M 35mm f/1.4 ASPH.
• 35.0 mm
• 1/350 sec
• Center-Weighted Average
• Manual exposure
• 1
• ISO 320

"Not calibrated for watts/steradian" (humor). I've not done that since 1990. But I was offered my old job back not long ago.

 

[Brian]

I'll have to try this by coding it up.

The spectral response of the CCD used in the M9 is "about" like most cameras. You can see the overlap of the three colors. And I am reminded of my Dad's old joke picking out a white shirt. "But I wanted a Green Shirt". Salesman reaches around behind and turns on the green light. SO- with a subject that is Green: Blue and Red still respond- because they are sensitive in green. Take the response shown for the portion of Green that they are sensitive to, and Boost it to Equalize with Green. Same with Red- Boost it to the green response, because you KNOW it is responding to green. Now you get a better monochrome. SO the algorithm: Use the 4 pixels in each Bayer cell to determine the predominant color. If it is Green- boost Blue and Red to the Green response. If it is Blue- Boost Green and "maybe red" depending on the specifics. I'd leave red alone based on the above plot. If it is Red- boost the green and blue. The assumption is the Bayer cell has an even color coming across it. It's the Green light on a white shirt- you know the Green determines the color. The Bayer cell is the White Shirt.

I wonder if any monochrome conversion software does it this way? It would have to know the spectral response of the camera, meaning good information on the color profile.

 

[PentUp]

Just a couple of very quick conversions from colour to B&W using free microsoft photo editor. Dialed out the colour, lifted the "clarity" reduced some light and slapped on the "vanilla" filter effect. Took seconds to do... so the result may be rubbish, but what the hey

2018, Pentax K5II & Sigma 50-500 DG OS HSM:

2014, Pentax K5II & Sigma 50-500 DG OS HSM:

 

[Dino26323]

Just a couple of very quick conversions from colour to B&W using free microsoft photo editor. Dialed out the colour, lifted the "clarity" reduced some light and slapped on the "vanilla" filter effect. Took seconds to do... so the result may be rubbish, but what the hey

2018, Pentax K5II & Sigma 50-500 DG OS HSM:
View attachment 9926

2014, Pentax K5II & Sigma 50-500 DG OS HSM:
View attachment 9927

Both look really amazing TBH. Really very good.
Both have light and shadow ( side lighting ) so yes black and White looks amazin here to me. Granted the colour ones may look good too. Cannot wait to get back and see them on my own screen. That is right. Black and White is at least in the race here.

Regards Dino26323

 

[Dino26323]

The black and whites are really well turned out here. My screen looks great with these two shots PENTUP. You are right to show the 2 conversions here or at least I believe so.