In the previous article, Nikon D3200 Long Exposure Noise, I ended up with the conclusion that the D3200 will produce a border of red noise that is visible during very long exposures. The question then was how to get rid of it.
1. Why Not Subtract It?
The obvious solution was to use some kind of dark frame subtraction and just subtract a noise frame from the photo. This doesn't work as well as one would hope, because unlike the noise that we subtract in normal dark frame subtraction, the noisy pixels do not remain the same. Dark frame subtraction corrects for variation in sensor pixels where some saturate too soon, while this noise is shot noise caused by radiation from the surrounding circuitry. While it is possible to use subtraction to add a general "anti-noise" to the border region of the photo, this has the effect of only approximately cancelling out the red noise at the cost of adding even more noise.
I tried a couple of times, but no matter how I played around with it I couldn't get a satisfactory result. It was never anywhere near the kind of automatic process dark frame subtraction usually is.
2. Crop It
In the end, I went with the most brutal solution: doing a center crop of the noisy photo. Since the noise is concentrated to the edges, removing them means that we can get rid of a lot of noise without (hopefully) losing too much image. The question then was how much to crop? What was the tradeoff between remaining pixels and remaining noise? To find out, I shot a dark frame
Then, for various crop factors, I summed the amount of noise left. For example, with a crop factor of two, you end up with an image one-fourth as large, but with very little noise. This graph shows how the average noise per unit area over the remaining image changes as we take a smaller and smaller center crop, with 100% being the average noise per unit area for the un-cropped image:
Note the red line in the plot. The amount of noise per area bottoms out at 84.9%, and if we set that as our minimum noise level, then the red line show us how close to the minimum level we are.
Here are the raw numbers. This first table is what was used to produce the graph immediately above. You can see, for example, that to get halfway to the minimum noise level, you should use a crop factor of about 1.15 - 1.20.
|Crop Factor||Noise / Area (%)||Noise Reduction (%)|
While the above may be interesting, I think this table is the one to use for figuring out your personal sweet spot. It shows the crop factors, the noise reduction as a percentage of what is achievable, and how many megapixels you get to keep. Here you can see that with a crop factor of 1.2, you get a noise reduction of 56.3% and keep 16.7 megapixels; which gets you a resolution of 5013 by 3333 pixels.
|Crop Factor||Noise Reduction (%)||Megapixels||Width||Height|
4. How to Use This
Once you have decided on the crop factor, use the lens zoom (if you have a zoom lens) to "crop" the scene for framing. For example, if you have a 10-20mm lens and have decided on a crop factor of 1.2: set it to 12mm, compose your shot, then zoom out to 10mm to capture. Then you crop it back down to 12mm in post processing.