White balance with filters - less noise in blue?

[cmi]

Hi gang,

Im photographing landscape and now I have some motives where I want to go again and re-shoot it in the highes possible quality. For one of these pictures my blue channel is really... horrible... so Im looking for optimisations naturally. I have read somewhere (cant find it!) that if you use colored filters for whitebalance, the quality of usually noisy channels like blue (quite often more noise), is enhanced. Is this complete nonsense or is it true? If yes, anyone has experience with this? What additional equipment I would need and how would I do it? (For one part I think it cant be true because then it would be a common advice.)

Also on a sidenote, why is it that the blue channel often gets more noise? I figure because of the different distributed wavelenghts in the particular spectrum, and if a channel is low it simply gets noise. But I dont know and I wonder if anyone could explain it with more detail.

Now I could just buy some filters and try for myself, and will probably just do that, but Im curious!

Regards,


Christian

[bjanes]

Hi gang,

Im photographing landscape and now I have some motives where I want to go again and re-shoot it in the highes possible quality. For one of these pictures my blue channel is really... horrible... so Im looking for optimisations naturally. I have read somewhere (cant find it!) that if you use colored filters for whitebalance, the quality of usually noisy channels like blue (quite often more noise), is enhanced. Is this complete nonsense or is it true? If yes, anyone has experience with this? What additional equipment I would need and how would I do it? (For one part I think it cant be true because then it would be a common advice.)

Also on a sidenote, why is it that the blue channel often gets more noise? I figure because of the different distributed wavelenghts in the particular spectrum, and if a channel is low it simply gets noise. But I dont know and I wonder if anyone could explain it with more detail.

Now I could just buy some filters and try for myself, and will probably just do that, but Im curious!

Regards,


Christian

Iliah Borg covers this topic and answers your questions. A magenta filter does the trick.

Bill

[cmi]

Many thanks for the link Bill, the whole site is great, Im reading it all right now. Phantastic infos there, can only recommend it. Looks like I gonna buy some filters. Oh and if anyone has more infos, please dont be shy to post

Christian

Iliah Borg covers this topic and answers your questions. A magenta filter does the trick.

Bill

[Panopeeper]

1. This technique is not "white balancing"; in fact it is a "white unbalancing". It is "optimizing the raw channel exposure".

The point is, that in most cases one raw channel determines the dynamic range, as pixels of that channel will clip far ahead of the other channels. Thus, if you want to prevent clipping, you have to keep the exposure low enough relative to that channel, which means the exposure of the other channels suffers.

In landscaping, the green channel is typically ahead of the others.

See following *raw* histograms:

- the first one is w/o filter; the green-red proportion is 13900:6100 ~2.28, i.e. the green is more than a stop ahead of the red;

- the second one is with a Tiffany CC30M; the green-red proportion is 11500:6200 ~1.85.

The difference in the proportions is ~0.43, close to one half of a stop.





The situation is very different for example under non-halogene incandescent lighting: the red will be dominant and the blue very low, so youl would need a cyan filter.

2. The downsides:

a. The white balance will be off. You have to preset it or have something in the image to pick WB on.

b. Filter quality: we usually give lots of money for high-quality filters for high-resolution sensors to prevent image degradation. The color correction filters are not available in the same quality; there is no multi-coating to prevent flare etc. B+W even discontinued the magenta filters.

c. Exposure: the filters block some of all wavelengths, but to different degree. The effect is, that the exposure needs to be increased more than the gain will be between teh channels. In the above example, the exposure increase was 2/3 EV, in exchange for less than half stop gain.

[cmi]

Yes, normalizing raw channel exposure at capturetime would be more appropriate as a correct term. Proper whitebalance would not be that much of an issue as long as you have a reference. But you bring up the topic of efficiency. That is one thing I have wondered about myself. You would naturally have a substancial improvement, and ideally the same exposure for all channels. ...a small filtercollection to get optimal exposure. Maybe with a polfilter on top, and suddendly you are getting into too long shutterspeeds. ...a bracketed shot might get rid of noise more easily.

But I will just try it. Seems like an option to get away with just *one* shoot in some situations. Guess I'll take one of the stronger mangenta filters for a start or maybe the Cokin filter system...

Christian

1. This technique is not "white balancing"; in fact it is a "white unbalancing". It is "optimizing the raw channel exposure".

The point is, that in most cases one raw channel determines the dynamic range, as pixels of that channel will clip far ahead of the other channels. Thus, if you want to prevent clipping, you have to keep the exposure low enough relative to that channel, which means the exposure of the other channels suffers.

In landscaping, the green channel is typically ahead of the others.

See following *raw* histograms:

- the first one is w/o filter; the green-red proportion is 13900:6100 ~2.28, i.e. the green is more than a stop ahead of the red;

- the second one is with a Tiffany CC30M; the green-red proportion is 11500:6200 ~1.85.

The difference in the proportions is ~0.43, close to one half of a stop.

...

The situation is very different for example under non-halogene incandescent lighting: the red will be dominant and the blue very low, so youl would need a cyan filter.

2. The downsides:

a. The white balance will be off. You have to preset it or have something in the image to pick WB on.

b. Filter quality: we usually give lots of money for high-quality filters for high-resolution sensors to prevent image degradation. The color correction filters are not available in the same quality; there is no multi-coating to prevent flare etc. B+W even discontinued the magenta filters.

c. Exposure: the filters block some of all wavelengths, but to different degree. The effect is, that the exposure needs to be increased more than the gain will be between teh channels. In the above example, the exposure increase was 2/3 EV, in exchange for less than half stop gain.

[Panopeeper]

Another issue: you mentioned the blue channel as the noisy one. This is suprizing, for in landscaping it is usually the red one, which is the lowest. I wonder what kind of scenery you are photographing, that the blue is worse than the red.

[bjanes]

Yes, normalizing raw channel exposure at capturetime would be more appropriate as a correct term. Proper whitebalance would not be that much of an issue as long as you have a reference. But you bring up the topic of efficiency. That is one thing I have wondered about myself. You would naturally have a substancial improvement, and ideally the same exposure for all channels. ...a small filtercollection to get optimal exposure. Maybe with a polfilter on top, and suddendly you are getting into too long shutterspeeds. ...a bracketed shot might get rid of noise more easily.

But I will just try it. Seems like an option to get away with just *one* shoot in some situations. Guess I'll take one of the stronger mangenta filters for a start or maybe the Cokin filter system...

Christian

Whether or not the exercise is worth it depends on your personal preferences and the noise characteristics of your camera. Gabor has already pointed out the down sides to this approach. You didn't say what camera you are using, but I presume that you are shooting at base ISO and using a tripod for your landscape work. Under these conditions, noise may not be a problem. Remember that shot noise is likely predominant in your images and it varies as the square root of exposure. Thus, if you could gain an extra stop of exposure in the blue channel, noise would only be reduced by a factor of 1/1.4. Emil Martinec has an excellent post on noise.

For testing, I would recommend Gabor's Rawnalize to visualize the raw channels. Remember that each 0.3 density unit in the filter reduces exposure by 1 stop, and 0.1 density unit decreases the exposure by 1/3 stop.

I have thought about using this approach, but with my Nikon D3, noise is not usually a problem at base ISO. Resolution of this camera is less than ideal for landscape, but by stitching together several images, excellent results can be obtained.

Bil

[Daniel Browning]

2. The downsides:

Another downside that Joseph Wisniewski encountered is that it can affect the colors, especially oranges, and a bit less in the aquas. The filter spectral response differs from the sensor CFA, so that where it crosses over from its red to green plateaus and green to blue plateaus often misses the wavelengths where the camera's CFA filters are crossing over. If there were lens filters that precisely matched the spectral response of the sensor's color filters, this wouldn't be an issue.

[douglasf13]

Another downside that Joseph Wisniewski encountered is that it can affect the colors, especially oranges, and a bit less in the aquas. The filter spectral response differs from the sensor CFA, so that where it crosses over from its red to green plateaus and green to blue plateaus often misses the wavelengths where the camera's CFA filters are crossing over. If there were lens filters that precisely matched the spectral response of the sensor's color filters, this wouldn't be an issue.

  Good point.  I wonder if using cameras with better color separation, like MFDB or A900, makes that problem less noticeable?

[Panopeeper]

The filter spectral response differs from the sensor CFA, so that where it crosses over from its red to green plateaus and green to blue plateaus often misses the wavelengths where the camera's CFA filters are crossing over

I think this could be countered with filter-specific conversion from the camera's color space in the raw converter's working color space - but who would make that effort for the single-digit number of photographers, who are using color correction filters to enhance the dynamic range?

If there were lens filters that precisely matched the spectral response of the sensor's color filters, this wouldn't be an issue.

How would *one* filter match the characteristics of *three different filters*?

[bjanes]

Another downside that Joseph Wisniewski encountered is that it can affect the colors, especially oranges, and a bit less in the aquas. The filter spectral response differs from the sensor CFA, so that where it crosses over from its red to green plateaus and green to blue plateaus often misses the wavelengths where the camera's CFA filters are crossing over. If there were lens filters that precisely matched the spectral response of the sensor's color filters, this wouldn't be an issue.

Joseph Wisniewski really knows his stuff. Thanks for posting a reference to that thread, which I missed the first time around. The Neodymium Enhancing Filter (didymium) filter for fall landscapes sounds interesting, but where do you get them? Back in the days when we did photomicroscopy with Kodachrome the didymium filter was widely used to bring out the eosin (red stain) in pathology specimens.

Bill

[Panopeeper]

I know this has no direct validity regarding magenta filters of other makes, but it might be interesting to take a view of the B+W CC Magenta filters' transmission characteristics:

[cmi]

Wow, many answers, and very interesting stuff!

At first, its a pleasure to gather all the knowledge here, VERY interesting facts and links.

bjanes, you are right, tripod + base ISO ( 400D). What you say about the efficiency of a colorfilter makes sense, one stop improvement expressed in noise is indeed not THAT much from my own experience. ...but these downsides of course wont stop me from trying it, I always have to find it out for myself. And by the way I have been working with Rawnalyze since you provided the link to Illiahs site. I must spend some time to learn it properly, at the moment Im just glancing over all this new stuff. And especially thanks for the density hint, as you may have guessed Im new into these topics, so such seemingly "simple" infos are very valuable for me.

Panopeeper, its interesting you note that I could counter the spectral response in post. As interesting as this is, Im not at this point now, but when I am, I'd be glad to come back to this. And yes, the red channel is noisy too... (in fact also the green in the particular image wich brought me to the idea) ...but it was not important to me to say exactly the channels, for me it was about general doability and not my specific picture... and the subtleties one has always to figure out by himself anyway.
And on a sidenote, the style you explain things is a bit too complicated for me. E.g. the Red/Green proportion you mentioned with some numbers made no sense (again, to me). So not to say I shouldnt learn about it, but for me it was just not clear waht you wanted to say. I can do some other things, but here Im an absolute newbie. For me, my experience tells me, generally the the best explanations to really get into a topic are these wich also my grandmother would understand.   For example the last table, I dont know if you intended to post it for me. I smell as a pro I should instantly recognize something very important from it, but I dont. Can you explain it? What exactly do you find interesting about that spectral response curve, and how do you interpret it?

Christian

Whether or not the exercise is worth it depends on your personal preferences and the noise characteristics of your camera. Gabor has already pointed out the down sides to this approach. You didn't say what camera you are using, but I presume that you are shooting at base ISO and using a tripod for your landscape work. Under these conditions, noise may not be a problem. Remember that shot noise is likely predominant in your images and it varies as the square root of exposure. Thus, if you could gain an extra stop of exposure in the blue channel, noise would only be reduced by a factor of 1/1.4. Emil Martinec has an excellent post on noise.

For testing, I would recommend Gabor's Rawnalize to visualize the raw channels. Remember that each 0.3 density unit in the filter reduces exposure by 1 stop, and 0.1 density unit decreases the exposure by 1/3 stop.

I have thought about using this approach, but with my Nikon D3, noise is not usually a problem at base ISO. Resolution of this camera is less than ideal for landscape, but by stitching together several images, excellent results can be obtained.

Bil

[cmi]

I think I answered the graph for myself, now that I slept about it. First I wonder why it is cut off at the sides. Is the diagram incomplete or is it really a "window" from 400-700nm abruptely coming in at around 90% density? For me it suggests that red and blue are also uniformly affected, even if they get only 10% more dark. Anyway, as far as I see, the curve illustrates the statement that was made by Joseph Wisniewski at DPreview regarding the tonal transitions between the primaries: It is unlikey that the camera sensor filter(s) have the same response, thus the color inaccuracies described by Wisniewski emerge.

Christian

I know this has no direct validity regarding magenta filters of other makes, but it might be interesting to take a view of the B+W CC Magenta filters' transmission characteristics:

[Daniel Browning]

How would *one* filter match the characteristics of *three different filters*?

Good point. What we really need is filter wheel at every pixel site, with dozens of selectable filter options. Right after they finish building antigravity suits and world peace.

For what it's worth, Iliah Borg respectfully disagrees with Joseph on the color accuracy downside. Personally, I haven't run into it either, so I continue to use magenta filters.

The Neodymium Enhancing Filter (didymium) filter for fall landscapes sounds interesting, but where do you get them?

B+W, Tiffen, Hoya, and Singh-Ray make them. They're generally called "red enhancing" or just "enhancing" filters, but the "contains rare earth elements" should give it away. (Hopefully they are referring to didymium and not radioactive Thorium  ). Here's a few links:

Singh Ray 67mm at their own site ($210 [!] for 67mm)
Enhancing and Intensifier category at B&H
B+W 67mm at B&H ($72)
Tiffen 67mm at B&H ($70)
Hoya 67mm at B&H ($63)

[bjanes]

Good point. What we really need is filter wheel at every pixel site, with dozens of selectable filter options. Right after they finish building antigravity suits and world peace.

For what it's worth, Iliah Borg respectfully disagrees with Joseph on the color accuracy downside. Personally, I haven't run into it either, so I continue to use magenta filters.



B+W, Tiffen, Hoya, and Singh-Ray make them. They're generally called "red enhancing" or just "enhancing" filters, but the "contains rare earth elements" should give it away. (Hopefully they are referring to didymium and not radioactive Thorium  ). Here's a few links:

Singh Ray 67mm at their own site ($210 [!] for 67mm)
Enhancing and Intensifier category at B&H
B+W 67mm at B&H ($72)
Tiffen 67mm at B&H ($70)
Hoya 67mm at B&H ($63)

Daniel,

Many thanks for the additional references. Perusal of the B+W and Hoya literature confirms that these are didymium filters. The Singh Ray site does not specify what is in their enhancing filters, but states the following:

"Years ago, when color enhancing filters were introduced, they were made from a mix of rare-earth elements known as "didymium" glass. These filters accentuated red and orange areas in foliage, sunsets, and similarly warm subjects. Often, however, they would produce color transparencies marked by an overall magenta cast in the neutral, white, green and lighter pastel shades of such scenes. Many photographers and picture editors found this unnatural magenta cast objectionable and unacceptable.

To solve this problem, the Singh-Ray Color Intensifier was introduced in 1995. While it does not "pump" the red in a scene quite as much as didymium filters, the overall results are much more natural and believable (in part because green foliage is also enhanced). The cleaner rendering of neutral colors and white areas are big reasons for the popularity of Singh-Ray's Intensifier Filter among serious outdoor photographers."

The price of the Singh-Ray is too high to buy one for testing without prior research. Do you have any experience with it? It is claimed to be made for digital, but this claim is often made for products that have no special advantage over older "analog" products made for film.

Bill

[Panopeeper]

First I wonder why it is cut off at the sides. Is the diagram incomplete or is it really a "window" from 400-700nm abruptely coming in at around 90% density?

This is the spectrum of visible light.

For me it suggests that red and blue are also uniformly affected, even if they get only 10% more dark

The magenta filter is the most "regular" in trasmission. Compare it for example to the red (below).

as far as I see, the curve illustrates the statement that was made by Joseph Wisniewski at DPreview regarding the tonal transitions between the primaries: It is unlikey that the camera sensor filter(s) have the same response, thus the color inaccuracies described by Wisniewski emerge.

I see this very differently. There is no "plateu" here, and the effect has IMO nothing to do with what he wrote.

For me it is a problem of white balancing: the filtered light as illumination is far from the black body radiation, it can not be expressed in terms of temperature and tint, therefor it requires a filter-specific color space conversion.

Btw, I don't understand what Joseph Wisniewski wanted to communicate. He sees color reproduction problems on one side, and suggests the Redhancer filter on the other side. Take a look at the transmission of the B+W Redhancer. The result may be fancy but it has nothing to do with correct color reproduction (though there is nothing wrong with wanting "pleasing" colors instead of "correct" colors).

[archivue]

the answer change according to the sensor you are using... sometimes, using filters doesn't help at all, other times it helps...

[cmi]

Could you elaborate on this?

... the filtered light as illumination is far from the black body radiation, it can not be expressed in terms of temperature and tint, therefor it requires a filter-specific color space conversion....

[Panopeeper]

Could you elaborate on this?

I should not have mentioned it :-)

Seriously, this is not a small topic (nor is it on-topic). You can find all possible and more information in internet by searching for "black body radiation". The essence is, that the spectrum of radiation of the sun and of incandescent (i.e. glowing) light sources is near-continuous.

See the solar spectrum (it is not so clean on the earth level, but the visible part is still quite continuous).

If the illumination is modified by such a filter as the Redhancer, the originally continuous spectrum becomes "ragged"; this illumination can not be described the conventional way via "temperature" and "tint" (some fluorescent sources too create this problem). The only way I know of to counter this would be to modify the color conversion from the camera's color space in sRGB, etc., specifically customized to the filter.

This is theoretically possible with the Adobe DNG Profile Editor (note: it is not only for DNG raw files!), but I don't think the effort is justifiable.