Originally posted by clackers You can't increase noise by cropping, because noise is a ratio, Class A.
You don't crop much, do you?
If you did, you would have noticed that heavy cropping can bring out image noise that otherwise remains hidden.
Originally posted by clackers If every twentieth pixel is noise, that is true inside the cropped area, and out.
It doesn't work that way.
As a matter of fact, what we are interested in is signal-to-noise ratio (SNR). We can have a ton of noise but if we have a huge signal as well, the noise won't matter. Conversely, a small noise level can be very visible, if the signal level is low.
By cropping, you reduce the signal (after all, you crop out a lot of signal). That signal (and its noise) is no longer available to raise the signal level and help noisy pixels even out each other. As long as you have a lot of pixels contributing to a square mm of the final image, the individual pixel noise will cancel each other out (to some degree). The fewer pixels contribute, the more noisy that square mm in the final image will be.
If you do not believe your own eyes (when cropping) then perhaps
the math in the article "Contrary to conventional wisdom, higher resolution actually compensates for noise" may convince you. See the formula on page 2 of the article how downsampling (using many pixels to calculate the values for fewer pixels) increases the SNR of the image (i.e., reduces its noise). The formula on the last page tells you exactly how much noise advantage you can expect from downsampling.
You are correct in stating that the same f-ratio leads to the same exposure, independently of the sensor size. That's correct, because exposure measures light intensity (e.g., lumen per square mm), which is independent of sensor size.
Let's say we are using the same pixels (same size and capacity) for an APS-C and an FF sensor. Then we are just using more pixels for the FF sensor in a larger area, right? If we now record the same scene with both sensors and achieve the same exposure (that means we are using the same f-ratio) then at pixel level we have the same noise (as exposure was the same). But when we now use all the many more FF pixels to downsample them to an image that has as many pixels as the one from the APS-C sensor, we increase the SNR. By a bit more than a stop (1.17 stops, in the case of Pentax APS-C).
To counteract this increase in SNR, we can stop down the FF lens by 1.17 stops, which by the way also establishes the same DOF as the APS-C lens produced. We then have an "equivalent" image with the same scene capture, same DOF, same noise levels. Even though the FF capture used a lower exposure, it doesn't matter, because we don't have to enlarge the FF image as much as the APS-C (which is equivalent to downsampling the FF image to APS-C dimensions).
Hopefully you can confirm that enlarging images increases noise. Every darkroom owner knows this.
Originally posted by clackers Bizarre.
Where did you read this?
I did not read it anywhere. It makes sense intuitively. If you take four pixels that can be either on or off then you can capture five different overall image brightness levels. If you only have two pixels then you can capture only three different image brightness levels.
If our pixels always define the same brightness (i.e., our sensor sensels always have the same full-well capacity) then this increase in image brightness levels does not result in a finer resolution of brightness levels, but in an increase of dynamic range.
Remember that by projecting the same scene to a larger area (with the respective amount of more pixels), you reduce exposure, i.e., a single pixel now receives less light then before and we can use this gained headroom to capture more dynamic range in the scene.
Also, if you like another source: Ever wondered why DxOMark dynamic range measurements seem to be too good to be true, i.e., specifying more DR than the bit-range of the camera seems to support? The reason is that
DxOMark state the DR in terms of a normalised 8MP sensor. If a camera has more than 8MP, it receives a respective increase in DR. The converse is also true, i.e., the DR for a 6MP camera would be stated lower than it is (for 6MP).
Last edited by Class A; 07-14-2015 at 06:28 AM.