Re: Guide to Zero Noise with Median Stacking - Photography & Photoshop CC Tutorial
As a scientist working with statistic and signal processing, I though I might add a few comments to this interesting thread.
Firstly,
@wornish, if I understand your example correctly, it is rather misleading. The top picture is a single exposure of 1/15 s, correct? The bottom picture is a stack of five exposures, so it contains information from 5 * 1/15 s = 1/3 s, five times longer. Regardless of the median stacking, the bottom picture has a much longer effective exposure, so, obviously, the noise will be lower. But perhaps I misunderstood your example.
I think the important question here is the one asked by
@Geoffc. Consider five shots of 1/15 s and one shot of 1/3 s, but with lower ISO. They should have collected the same amount of photons, so why would the median stack resulted in lower noise?
This is because the median is not sensitive to outliers, i.e., speckles or hot pixels. If you open your shutter for 1/3 s and collect all the light, there will be photons with unusually high energy coming occasionally, creating what we see in the image as speckles. If you divide this exposure into five, the chances are that in the given pixel the speckle event will happen only once, maybe twice. The median will help eliminating these "outliers".
Imagine you observe one pixel in one colour channel. Let us assume that the typical value of this pixel in this channel is 65 (you could establish this taking a long exposure at very low ISO). Now, let's take five short pictures. We can get values like these:
65, 68, 61,
95, 64
The numbers marked in black represent typical noise, while the 95 in red is a spurious event. Now, if you took one longer exposure instead, the result would be the
mean of these numbers, about 71, which is brighter than the typical value of 65. If you take the
median of these five numbers, the result will be 65, close to a typical value.
The median stacking can help you
lower some types of noise, namely speckles and hot pixels. It will not eliminate this type of noise entirely, it will not eliminate background Poisson noise, thermal noise and other sensor artefacts, it will not lead to zero noise, as claimed in the title of the thread. I'm sorry guys, but you cannot beat physics and statistics. Noise will be always present in your pictures.