Lately I took a set of pictures with night urban scenes. Such a subject requires relatively long exposures and ISO should be in most cases high enough. All this stimulates noise and so, normally night photos contain a good deal of noise artefacts. To supress noise I applied some techinque that I took from my astrophoto experience and now I would like to speak on this matter.
So, for all my night shots I used tripod and took several shots from the same point and with all the same sets (typically from 5 to 10). Surely, I saved pictures in RAW format since (a) it does not use any compression and so, it does not rape any information; (b) it uses 13 bits to code brightness (for my Canon 350d) in contrast to JPEG which does only 8. 8 bits are quite enough for resulting image but it maight be not enough at post-process stage, e.g. if one needs to stretch histogram significantly. For cathedral image at the left I took 5 shots (ISO400, f/16, t=2 sec).
I start preprocessing with converting my 5 RAW images to 16-bit TIFFs correcting color temperature and exposure if needed (it goes without saying that corrections must be the same for all 5 pictures). TIFF is not crucial, any 16-bit format would be OK.
In PhotoShop I open new 16-bit canvas with the same sizes that my 5 images are of (in my case it is 3456x2304px). I put these 5 images into new canvas as 5 layers. Now the main trick. I set transparecies for each layer by some specific way (using 'normal' mode): 100% for lowest layer, then 50% for layer that is just above the lowest one, 25% for next one etc... For my case with 5 images transparency values will be as following (from top to bottom): 6.25%, 12.5%, 25%, 50%, 100%. Then I join all layers into the only one.
With such set of transparencies when each previous layer is with factor 2x more transparent comparing to the next one
, the brightness value for each pixel on resulting ('stacked') image represents a mean value for 5 images that participated in stacking. With such stacking real objects which are presented on all 5 shots and which are the same on all 5 shots (sky, clouds, buildings) do not change, while noise (which is result of spontaneous dark current and is not the same on each of 5 shots) becomes lower.
It is illustrated by the picture below on which part of single shot with sky (that typically is the most noisly place) is shown at the left with 200% magnification to show more details and result of stacking procedure (as described above) is shown art the right with the same magfnification. Here is link to this picture
Essential improvement for SIGNAL/NOISE ratio is clearly visible. Note, that with 10 shots effect would be even more remarkable. From my experience I may conclude that sometimes even 2 images (instead of 1) are enough for successfull noise supression. In most cases I use 4 or 5 and it works properly. 10 images were enough in all cases I faced with.
I would be glad if my post will be of any use for somebody.