Need a little help writing Photo 101.7 ISO

[neilcreek]

Hi folks!

I'm writing the newst Photography 101 post for DPS, on the topic of ISO. Gee wiz this is a tricky one. ISO is a monstrously complex and esoteric subject, and I'm struggling to boil it down to the simple facts, present them in a clear and easily understood way. Illustrating the concepts is also tricky, and I'm not 100% sure I understand the basics myself, so I was hoping I could run some questions by you all to help me make sure I'm getting it right.

Here's an excerpt from a relevent section talking about how ISO works. Does this sound accurate to you?

...we can increase the ISO setting on the camera. When we do so, we are increasing the signal gain. Essentially this is like turning up the volume. All of the values of the measured current (whether from light or noise) are increased. Each doubling of the ISO value, is a doubling of the gain: a doubling of the measured current in the chip.

Doubling the light in your photo is a pretty easy way to make an otherwise under exposed photo bright enough, but it comes at a cost: you lower the signal to noise ratio, and the noise becomes more apparent.

I understand and can clearly see that noise becomes readily apparent in a dar part of a picture, where the S:N ratio is very low. However why do we still see noise in large bright flat areas like the sky? The signal is many times greater than the noise, yet it can be distractingly obvious. Is the noise added (and subtracted, depending on the random variation) to the value of the pixel?

I'm also struggling to show the effect of noise on a graph vs signal strength, and how the noise profile changes as the ISO (gain) is increased. Is only the amplitude of the noise increased, or the average value, or both?

Thanks for helping me get this straight! If I'm not clear on the details, there's no way that I can hope the readers will be

[Jim Bryant]

One way to improve sharpness in dim light is to increase the camera's sensitivity. This works in places such as theaters and gyms where subjects are too far away for flash to work and where you need a faster shutter speed to eliminate blur. It also is a good way to get pictures without using flash in places such as concerts and museums where flash is prohibited.

Sensitivity is usually specified as an ISO setting just as the speed of film is. Increasing the camera's sensitivity or ISO means less light is needed for a picture so you can use a faster shutter speed to freeze action or reduce blur caused by camera movement. Sensitivity on some cameras can be set between 50 and 6400, a range of 8 stops, but most offer a smaller range of settings. The price you pay for using the higher settings is noise—randomly spaced bright pixels concentrated in dark areas of the image. The more you increase sensitivity, the more noise you get. This is because digital cameras increase sensitivity by amplifying the signals captured by the photosites on the sensor—similar to turning up the volume on the radio. Dim light can be made brighter this way but unfortunately, amplifying the image also amplifies noise. Many cameras have a noise reduction mode designed to reduce or eliminate noise caused by long exposures. Some allow you to turn this mode on and off.

[OsmosisStudios]

Hold on a second: youre writing an article for the mainpage and asking us about the basics of the concept?

Sorry to say, but I hardly think you should be writing the article at all.

[neilcreek]

No, I'm asking for some knowledgable experts to confirm my thinking and clarify a couple of points I'm unsure about, to make sure that I get it right for the post. Would you rather I publish inaccurate posts? Or are you demanding that all DPS authors be experts in all fields? We're part of a community here, and I thought it was appropriate to involve others in helping make the best post possible.

[Doug Pardee]

The reason that you see noise in the sky even at low ISOs is basically because the amount of "signal" present in the blue channel is inherently low. White light is only about 7% blue. The noise generated in the sensor element, however, is the same for all of the color channels. So there's a lower signal-to-noise ratio for blue.

However, that is not the entire story. The sensor data is not used "as is", but is adjusted for white balance. Typically the blue sensor data is multiplied by maybe 125% compared with the green channel, while the red sensor data might be multiplied by 250% compared with the green channel.

Working backwards, we get:
• Green: 71.5% of white light, not multiplied (our reference level)
• Red: 21.3% of white light, multiplied 250% -> sensor data levels were 11.9% of green
• Blue: 7.2% of white light, multiplied 125% -> sensor data levels were 8.1% of green

So we see that Red will have about 3 stops more noise than Green, while Blue will have about 3-2/3 stops more noise than Green.

The Blue noise tends to be most evident in skies. The Red noise tends to be most evident in skin tones (Caucasian) and is particularly noticeable when converted to monochrome (black and white).

Edited to add: oftentimes what one notices in the skies is not so much noise in the Blue channel, but rather Red channel noise due to the low signal level of Red in blue skies along with the relatively high multiplication factor for Red. Blue noise in skies manifests as luminance noise (blue "grain"), while Red noise in skies manifests as chrominance noise (red speckling).

[candleman]

i havn't read everything above.. heres my short blurb.

there are 3 types of noise.
sensor noise, colour noise, an chroma(luminance?) noise..
they do look different, but generically they all come about from sensor quality vs lighting, vs ISO sensitivity.
i believe "big boys" photoshop (cs3 etc) has indivital settings to remove each type,
i know Photoshop elements (in ACR too) has 2 sliders for noise reduction, one for luminance, one for colour.

thats all i have to say about that.. google it and you';ll be reading until Christmas

i just take it as a fact of life
high iso+underexposure/dark = yucky

therefore flashgun/wide apertures are on the prescription.

[OsmosisStudios]

Originally Posted by neilcreek

I'm not 100% sure I understand the basics myself, so I was hoping I could run some questions by you all to help me make sure I'm getting it right.

I seem to have struck a nerve, but you have to admit that it was somewhat justified.

Im not asking all DPS posts to be perfect, but when someone says they arent sure about the basics, it leads me to believe there may be others that are more qualified.

[neilcreek]

Thanks for the replies folks. Doug, that insight into the varying gain on different channels is both fascinating and frustrating, as it adds a further element that I had not been aware of. It's very interesting to know, and while I think it's beyond the scope of the Photography 101 post, it does help to solidify the concept better in my head. If it's okay with you, I may refer back to your post in the future when I return to write on the subject of noise in more detail.

Candleman, thanks for the mention of the different kinds of noise. I am aware of that through my research on the post, specifically this page on the fantastic Cambridge in Colour site: http://www.cambridgeincolour.com/tutorials/noise.htm

Osmosis, the nerve you struck was due to the tone of your post. I apologise if I was too rude, and in the light of the sentence you quote, your comment makes more sense. I misspoke when I said that, and actually meant that I need to sort out the details of the complex issues in my head. By "basics" I probably meant fundamentals, which in this case are by no means basic. I'm hoping once you read the article after it's done, you'll find it to be a valuable resource, worthy of being posted to DPS. Darren has had a read of my article and likes it a lot so far.

I'll finish a draft of a couple of the illustrations I want to use, and if I could run them by you all for accuracy and legibility, I would appreciate it!

[neilcreek]

Here's the first graph, simply showing noise vs signal in a low noise example. As the signal gets greater (vertical axis), the low constant average noise stays the same, and the signal to noise ratio stays the same. It's only in the darkest areas that the ratio falls low enough to be of concern:



I've seen a few graphs like this around the web. Is this accurate? One thing that confuses me: if the vertical axis is brightnes, what is the horizontal axis?

In this animation I've taken a cross-section of a radial gradient across one row of pixels as an example. Each frame assumes a halving of the actual signal, and a doubling of the ISO to compensate. That keeps the shape of the signal constant, but shows the difference in the noise.



Does the average noise value, or amplitude, or both (as I have shown here) increase? Is the graph a good representation of the change in S:N ratio as the ISO is increased?

[Jim Bryant]

good start, you should also shot it picture wise as well.......graphs are greek to some folks.