shorterimage

Shooting on the Mean Streets of Noise

Most people know that as you raise your camera's ISO the noise goes up proportionally. Double the ISO and you double the noise. Double the ISO again and you've quadrupled the noise. Therefore when you shoot in very low light, you generally need a tripod to get acceptable noise levels in your low light images. I say generally because there are lots of factors going on here as modifiers.

Different cameras have different noise spec's meaning some cameras will be able to shoot in lower light than others. Also, now due to the marvels of technology some cameras have image stabilization in the lens or in the camera body. And with the right combination you can actually have both, in lens and in body stabilization working together. This can give you up to 6 stops of stabilization given the right lens/camera combination.

What am I getting at here? There is a way to reduce photo image noise by using the light gathering capabilities of multiple exposures. You then combine them into one image using a Mean algorithm that keeps all identical data and discards all different data. This works because real content in the multi-image scenes will be the same and noise, which is random will be different in each shot and eliminated. That's the technique that is being employed here in this tests!

For the test done here I was using my Sony A7RII with in body sensor stabilization with an older lens that does not have stabilization.

All shots here were hand held! I did the test this way knowing that images would be sharper if shot on a tripod. But what I really wanted to test here along with the noise reduction was a quality assessment of doing the noise test in conjunction with hand held shooting. I wanted to see a real world test of what quality you can expect when hand holding long sequences of shots that will then be merged into one image.

The first thing we have here is a composite image of a Teddy Bear sitting in a chair in very low and indirect indoor light. This was a sequence of 11 images shot hand held at f8, 1/50th second exposure and ISO 3200. However I underexposed this on purpose by about two stops. Yeah, I know it would be a more defined test if I shot it two stops brighter by upping the ISO by two stops. However I underexpose a lot with my A7RII (long story why but it goes to tests I've done with the A7RII's BSI sensor and dynamic range). You suffer some noise in underexposing but the A7RII has a great noise spec and I think with the A7RII you get the optimum dynamic range by underexposing a stop or two.

Note 1: This process uses Lightroom and Photoshop. Other applications may work and provide the same functions but I have not tested any other than the applications used here.

Note 2: Subjects should not be moving to use this technique or else you will get blurring in the spots where there is movement!

Which brings up the prospects of what a landscape scene with moving water or sky might look like. I think I'll test that and see!!

Above in the top panel you have the original full frame shot on the left and the Mean Averaged noise reduced frame on the right. Then you have a 100% crop of each followed by a 200% crop and lastly on the bottom a 400% crop of both the original shot on the left and the Mean Averaged noise reduced shot on the right.

Pretty dramatic difference huh?Keep in mind this was shot at ISO 3200, however being about two stops underexposed the effective ISO is really more like 12800.

I thought I'd torture this a bit more though so I shot the sequence below at ISO setting 12800. Once again I underexposed the image sequence by about two stops so the effective ISO for the images below is more like ISO 51200.

I know some of you are wondering about my underexposing these. Well there's a bit more to it than just the A7RII BSI sensor dynamic range thing. I was thinking real world shooting here for a worst case scenario. So I picked scenes that put things right on the edge. On the edge meaning I didn't want to go any lower on shutter speed than 1/50th to 1/60th seconds for these long sequences. It's hard enough to image merging long sequences hand held but shooting slower than this would even exacerbate things. So I thought this would be a good all around test with my A7RII having great noise spec's to see what a worst case, slow shutter speed sequence and underexposed image would look like using the Mean Average Noise Reduction technique.

And this rendered pretty good results given the parameters which tells me that when able to shoot with a tripod, or at higher shutter speeds and more normally exposed, I should get outstanding results with this. As things are I think they are pretty good already.

The image sequence below was shot at f8, 1/60th second exposure and ISO 12800. First up are the full frame images, original on the left and Mean Average noise reduced on the right. In looking at the results carefully, what I see is greatly reduced noise on the right but also a little bit of softening and blurring which I assume is due to fairly slow shutter speed. Not quite as sharp but a much better image over all I think. It also tells me that given a little better shooting scenario using higher shutter speeds, or optimum on a tripod, and I should get very good results with this technique.

Original ISO 12800

Mean Averaged Noise Reduced ISO 12800

100% Crop

200% Crop

And here's how you do this noise reduction technique!

Obviously, shoot a sequence of images. You can calculate out how many images if you want your final Mean Averaged shot to meet a finite noise characteristic. I.e., If you want the final result to meet an ISO 100 noise spec, at ISO 200 you would need to shoot 2 images. For ISO 400 to meet an ISO 100 noise spec you would need to shoot 4 images. For ISO 800 to meet an ISO 100 noise spec you would need to shoot 8 images. And for ISO 1600 to meet an ISO 100 noise spec you would have to shoot 16 images.

Get it! For every full stop you double the ISO, double the previous full stops number of images you would need to shoot to meet your starting ISO noise spec. Since I didn't hold to that scenario here and also underexposed, my effective noise spec is probably more like ISO 800 for the Bear image sequence and probably up around ISO 1600 for the Christmas pottery image sequence.

What you do after shooting the sequence and importing them into your computer is you open them in Lightroom. Select the sequence of images. Right click on the selected images and in the pop open menu select Edit In, and then Open as Layers in Photoshop.

Below is the result you get once Lightroom sends the image layers into Photoshop.

After getting all the layer image sequence into Photoshop, you next select all of the layers (see all selected at right), then from the Edit menu select Auto Align Layers. If you are shooting on a tripod under optimum shake conditions you may not need to do this step but I think still probably wise to do. It simply aligns all like pixel content in each layer. See next step/image once the resulting menu pops up.

Selecting Auto Align Layers will open this menu. Keep the default selection of Auto and select OK. This will open a progress window that takes a little bit of time to process.

Once you've Auto Aligned your layers, if you hand held your shots you will probably have some edge errors to correct for. This is because when aligning layers of the sequence there was most likely some slight movement between each shot. In aligning them, slight differences for size and axis movement will be made resulting in some spill areas out at the edges. In this next step below, you can use the Crop tool (selected on the left) and slightly crop in your image to correct for the edge spill.

Next, open the quick menu select, the hashed menu selector to the right of the Layers panel, and then select Convert to Smart Object. Note all of the layers once again need to be selected for this conversion.

For the last step of the procedure, under the Layers menu, select Smart Opjects, Stack Mode, and then select Mean. This will kick off a process that will take a few moments. Longer depending upon how many images you have in your sequence.

Once this process completes, you will have your final noise reduced image.

At this point you could save it as a .psd if you want to keep the entire layered process. And of course you can also Flatten the image and save it as a final DNG, TIF, PNG, JPG, etc.

The most fun at this point is to look at the original and the noise reduced image side by side. You'll be blown away by the amount of noise reduction this technique renders!