Differential responses CCD/CMOS sensors vs the naked eye

[antonroland]

Hello all

Yet another question on tehnical detail but with a bearing on photographing evidence illuminated with alternate light sources.

I am fortunate to have time to experiment while photographing latents and found that colour reproduction and sensitivities are often vastly different from what I see to what is captured on gthe camera.

This becomes especially relevant when a print appears better when viewed through a red barrier filter but the photographed print is not as good as the inferior looking result viewed through an orange filter...

We use Pentax *ist DL D-SLR bodies for interest and I would like to know if others have experienced the same and what cameras you use. I am forming the idea that the sensitivities of those sensors (in my case a 6Mp CCD built by Sony) clip way inside the approximate visible spectrum of 400-700 Nm. As far as I have seen to date the clipping is greater on the UV to blue side than on the red to IR side.

Would love to hear your experience in this.

Have a good one!

[Gerald Clough]

I know CCD's have a higher red sensitivity with a sharp cut-off at about 700 and a distinct lack of the high human eye blue-green peak, while the human eye in daylight adaptation has a fall-off at the red end that begins about 600 and drops off entirely by 700. I know a lot of attention has been going into improving the blue-green response of CCD. Kodak talks a lot about it, and Sony added "emerald" to their filter array in place of about half the greens. And the dark adapted human eye really takes a beating on the red end. I would expect that to make the difference even more pronounced.

[antonroland]

Hello Gerald

Thanks for the response and trust you are well. Do you have any similar info on CMOS sensors? I know Sony have had the 4 colour CFA for quite some time so this is why.

I photographed some dusted prints through a red barrier filter recently as opposed to the orange we mostly use and I was quit unhappy with the apparent near burnt-out hotspots, something I never had with the orange barrier filter.

FWIW, we use a Rofin Polilight PL-500 and my favourite wavelengths are 450, 505 and 530. Now according to the book 505Nm is still red and orange barrier filter country but 530 and up towards red is red barrier filter country...hmmmmm great for viewing but not nice for photographing.

What about the UV - purple end?

So we learn I suppose, eh?

[Gerald Clough]

That just about plumbed the depths of what I know. But see if this helps:

In CMOS sensors, the blue peaks at about 480 nM versus 419 for the human eye. The green peaks at about 535 nM versus 531, a pretty close match. The red peaks at about 645 versus 559 for the eye. These differences are not so significant though, as the ranges of the colors.

We can see that the CMOS sensor will report color data that is invisible to the human eye, and this happens strongly in the infrared part of the spectrum. For this reason, an IR "cut filter" is routinely placed on the front of any consumer CCD or CMOS imaging device, be it camcorder or camera. This removes some of the extraneous image data and allows a more realistic image to be created. Now, see how the blue and green curves overlap? This means that something that is blue-green stimulates both sensors and they together produce signals that, when mixed, appear blue green in the resulting image. The same is true for red and green overlap- we see this as yellow. If a signal triggers both red and blue, we perceive the results at violet. A signal that is equally strong on all three appears as white.

Now, image sensors produce terrible color at the first approximation. Because the spread of the colors that each filter or dye responds to in the sensor or the eye is so broad, the colors tend to be poor in quality- washed out and weak in saturation. The "richness" of the image suffers. In the eye, our brain does the math to figure out the true saturation and how things should look. In the image sensors, a special math algorithm is used to create the proper colors by calculating the spectrum overlap and sorting it out to create the true saturated appearance as it would be to our eyes.

Designers have arrived at specific math transformations that will narrow the spectral data to certain colors, thereby creating the proper separation of the overlapping red, green, and blue data. These math processes are simple enough that they can be programmed right into the chip so that it happens automatically. This illustrates the point of this effort - the brain (and the digital camera) are taking a wide range of color information and distilling it down to data that fits in three and only three channels. Those three channels are the three primary colors used to create images, and even though some of the information is from colors that overlap a great deal, the bottom line is that you need only to "pigeonhole" that data to have what you need to create "true" color images.

So, it sounds like you need specific data on the particular sensor chip being used. And unless you're capturing in RAW and viewing without any color adjustment, you have to factor in the camera's white balance schemes. I agree with your notion of a filter good for viewing but poor for photography. Sensors are natively sensitive well into IR and are internally filtered and further massaged at the chip level, and nothing can variably track the eye-brain adaptive combination and variably shape the filtered sensor to match.

I saw some comparison of two cameras by a single manufacturer in which the two had dramatically different results with an R72 filter. Vast difference. And apparently manufacturers don't talk much about their IR internal filtering, since it goes to their various claims for capturing "natural" color. I assume they play with their IR cut filtering until they judge that it works with their sensor to please users, and some obviously still pass considerable IR. Adding a clear IR cut filter probably is futile, as that's essentially what they've done already. Maybe try a blue-green IR cut filter. It cuts off sharply at about 780. But that's assuming your camera's internal filter is passing enough IR to have an effect under the special conditions you're working under. You're not really looking for exactly a match to your human vision. You're looking for discrimination in the captured image, whether it appears similar colorwise or not.

[George Reis]

Anton - I think the issue you are seeing is from using a camera with a color mosiac filter - regardless of whether it's CCD or CMOS based. As such, depending on which wavelength(s) of light passes through the cut-off filter, you may only be capturing with one fourth to one half of the pixels on the imaging chip. If you use a monochromatic camera, you would see substantially better results with narrow band width photography.

If you are filling your frame with the latent, then I would expect that you are still getting good results (just not as great as you hoped). You can improve the results through post processing - especially with the channel mixer or the black and white adjustment in Photoshop.

Feel free to send me a sample image and I can take a look, if you like.

George

[antonroland]

Hi George, good to hear from you!

I have found a Tri-x B&W emulation action somewhere on the net and it gives me rather pleasing results. It is essentially a series of channel mixer adjustment layers and one or two other tricks so hopefully in line with your suggestion. It definitely yielded the best B&W conversions I have produced in P.S. to date...and I am full of, you know what, with B&W due to my film darkroom experience.

The newly discovered added bonus of this action is that it very nicely gets rid of hot spots in the red barrier filter captures.

You talk of monochromatic cameras...how would this differ from capturing in B&W mode with a regular D-SLR? I must see if our cameras are actually capable of doing so...

Will mail you later today.

Have a great Friday all!

[George Reis]

You talk of monochromatic cameras...how would this differ from capturing in B&W mode with a regular D-SLR? I must see if our cameras are actually capable of doing so...

Monochromatic cameras have no Bayer filter (or other color mosiac filter) covering the CCD/CMOS chip, so all wavelengths of light that the chip is sensitive to are received by every pixel. Color cameras with a black and white mode are simply post processing the image in much the same way you can do so using the channel mixer, but you still have that mosiac filter that is not recording data on up to 3/4 of the pixels when using strong color filters on the lens or illuminating with narrow bandwidths of light.

There is a way to demonstrate this using Adobe's DNG SDK. It requires using the command line interface on a raw file, but it will allow you to see a raw file without a gamma correction applied and without demosiacing applied. The image is grayscale, and shows how much light was recorded by each pixel.

George

[antonroland]

Is it possible to gut a regular D-SLR and remove the Bayer filter?

Good 2nd hand 10 Mp cameras go for cheap and that could work I suppose...

[George Reis]

I don't think the color mosiac filters can be removed - but if they could, it may be more expensive to make that modification than to buy a monochrome camera. Most monochrome cameras will need to be tethered to the computer as they generally do not have a viewfinder. A scanning camera would also be an excellent solution, as you have full color capture, but each pixel position records red, green, and blue. You may not get quite as good of a sensitivity as with a monochrome camera, but you would retain full color capture which is useful in cases with multi-color backgrounds.

George