high ISO

[Bill Pierce]

Smaller sensors, APS-C and Micro 4/3, are getting better and better, competing with full frame sensors in almost all departments - except high ISO. Bigger pixels mean less noise. An extreme example is the Sony A7s, 12 megapixels on a full frame sensor. High ISO is exceptional and blow ups held to 17x22 inch paper look good. (Remember, high pixel counts primary advantage is the ability to produce large prints.)

Don’t I wish that was true of the smaller cameras with their smaller pixels. These little fellows are ideal for discreet shooting in the available darkness until you want to make a big print from a file shot at the gear’s highest ISO. In the noisy old days of pixel peeping a lot of us used a program called Noise Ninja to reduce noise with a minimal loss of sharpness. Recently I have been using PhotoNinja, a raw processing program, because it does a good job with both Bayer and Fuji sensors. Noise control buried within that program is an updated version of our old friend Noise Ninja and taken off of its standard setting does an excellent job of dealing with extra-noisy files.

There are a lot of ways of dealing with the problems of low level available light - tripods, flash, fast lenses, high ISO. I use high ISO because it’s easy, just a twist of a camera dial and a little extra time with a processing program. But I wondered what techniques other photographers with small cameras used and what they felt were their advantages and disadvantages. If you’ve got a way that works, pass it on.

 

[willie_901]

I like your use of the phrase "noise control".

Noise (the uncertainty or error in the data) can not be reduced. You can not increase the information content (signal-to-noise ratio) of a single after the shutter closes.

All post-production noise control methods involve mathematical filtering which essentially reduces the information content of less noisy pixels to improve the appearance of nearby noisy pixels using clever averaging algorithms. Obviously optimal filtering parameters improve the images aesthetics. At the same time, we all know heavy-handed noise filtering parameters do more harm than good.

Selective noise filtering can be very useful, especially for large prints. I use the LR Adjustment Brush. There are numerous other post-production techniques that are probably more effective. Raw files respond better than JPEGs since raw files contain more information.

Digital photographers tend to think about noise, but actually our challenge is to optimize the signal-to-nose ratio. The signal level just as important as the noise level.

When time permits, the best approach to increase SNR is to increase the analog exposure of the sensor. Prof. Emil Martinec describes the process as:

• Set the aperture to obtain the desired depth-of-field
• Select a shutter speed that freezes either camera and, or subject motion as needed.
• Use the lowest ISO possible that's consistent with the exposure parameters determined in the previous steps. Electronic stabilization or a tripod provide more exposure-parameter flexibility.
• Finally, fine tune either aperture or shutter speed to retain only the interesting or important highlights in all three channels. This means some highlights may intentionally be overexposed. If every highlight region must be recorded, the shadow regions’ SNR will suffer.

This results in the highest possible analog SNR before the signal levels are increased via ISO amplification. Maximizing exposure minimizes the amount of noise filtering applied during post production.

Often spontaneity is more important than shadow noise and maximizing exposure. Content trumps SNR. Sometimes I bracket exposures in a burst and pick the optimal exposure in post-production.

Unfortunately different cameras use very different data path designs. With some cameras the analog SNR decreases significantly at a above a certain ISO value. Some cameras use different analog amplification circuits as ISO increases. This means shadow SNR may be optimum at above the sensor's base ISO even though the total exposure (signal level) is reduced. Almost all cameras use digital multiplication after the integer data leaves the ADC for extremely high ISO values. Some cameras don't use analog amplification at all as ISO increases.

To confuse things even more, some brands filter the raw data as well. A spatial Fourier transform of the raw data shows whether or not filtering is applied.

My cameras have a linear ISO vs SNR relationship. In very low light I set ISO to 200 (base ISO), record raw files and just use appropriate exposure parameters required for the scene at hand. Then I increase the global luminance (exposure) during post-production. The advantage for my cameras is the dynamic range is always as high as possible. The disadvantage is image review is usually not practical because the in-camera JPEG display is significantly under exposed. This technique would not be useful for other camera designs. Bill Claff has measured dynamic range vs ISO and shadow region improvement vs ISO for a large number of cameras. While these results involve dynamic range, a sensor's analog DR is a proportional to its analog SNR.