How clean can you get ISO 102,400?

In recent years the latest DSLRs have come with the ability to shoot at incredibly high ISOs.  The tradeoff of course when you shoot at these high ISOs is you get a lot of noise in the image too.  But you get the shot, so it’s the choice between having a noisy image and having nothing.  One of the things I’ve found interesting to test is how clean can you get high ISO shots.  The new Canon 5D Mark III will shoot at ISO 102,400, so how clean can you get an image taken at ISO 102,400?  Looks like pretty clean!

Before and After

Click to zoom

When processing my images I’ve always been interested in how clean I can get the image without losing detail.  I’ve tried all sorts of noise reduction software and they each have their pluses and minuses.  The very best way to reduce noise of course is to shoot at a low ISO.  But I’ve been interested in seeing how well I can reduce the noise in a high ISO image.

Over the years I’ve come up with various ways to process my images to reduce the noise in very high ISO shots.  When I got my new Canon 5D Mark III I had to test out ISO 102,400.  Straight out of the camera it’s pretty noisy.  But apply some noise reduction, reduce the image size, and it’s not horrible.  For getting a shot you wouldn’t have otherwise or having a picture for the web, it’s not that bad.

One way to reduce the noise in an image is to take an over exposed image, reduce the exposure in processing, and then combine that image with the properly exposed one.  The combined image will greatly reduce the noise in the shadows.  You can also combine/average multiple images together to increase the signal to noise ratio in order to reduce the noise.  Those techniques will be what I’m showing here.

So let’s see some samples!  Here’s a crop from an image taken at ISO 102,400 that was just converted to jpeg in Canon’s DPP.  It’s all default settings, no extra adjustments:

ISO 102,400

Click to zoom

The shadows have a lot of noise.  If you’re just looking for a snapshot, it can be used.  But anything else and the details are really gone.  I want you to notice two things to see the difference.  The sign in the background on the left isn’t even readable.  On the right side of the image is a building, and there’s nothing on it right?  It’s just black.  Here’s the same crop after doing a combination of exposures:

ISO 102,400 - exposure fusion

Click to zoom - noise reduction from exposure combination

Pretty big improvement huh? Notice the sign on the left, now you can actually make out the words.  The building on the right, there are three signs we couldn’t even see before.  Now we can see and read all the signs except for the times under the “OPEN”.  But we aren’t done yet!  This time, notice the solid colors of the carousel – the green, red, and yellow.  Let’s do a combination of exposures and averaging:

ISO 102,400 - exposure fusion and averaging

Click to zoom - noise reduction from exposure combination and averaging

Now check the solid colors between the last two.  They’ve gotten cleaner, less grit in them. We still can’t quite make out the times that the carousel is open though.

I’ve been working on a new method of combining the exposures and it’s so far looking very promising.  I already know that the new method brings out details in the very dark shadows well.  I’m still working on some kinks, because the solid colors aren’t as clean, but I know what I need to do to get them as clean as the last.  But here you can see the results of the new method, look at the open sign that we couldn’t read the times before:

ISO 102,400 - new exposure fusion method

Click to zoom - noise reduction from new exposure combination method

Now we can clearly read the times that the carousel is open!  But as you can see the solid colors aren’t as clean as before.  Here’s a comparison of those sections side by side.  The left is the original image, the center is with exposure combination and averaging, and the right is the new method I’m still working on.

ISO 102,400 Comparison

Click to zoom

And keep in mind, I’ve done NO noise reduction on the jpeg images, the raw converter’s noise reduction is set at it’s default setting for ISO 102,400 images.  That’s the only noise reduction that’s been done.  I’ve also done NO sharpening to these images.

The new method does bring out more details in the shadows, just need to work on it some more to make it better. While you can really clean up high iso shots doing this, it’s got limited uses.  Moving subjects wouldn’t work very well, since you are combining images.  There can also be a lot of processing involved to do this.  I’ve tried this on low ISO shots and while it does work, the benefits aren’t that great since the shots are already low in noise.

But it is interesting to see what you can do with an image shot at ISO 102,400!


Equipment used:
Canon 5D Mark III
Canon EF 70-200mm f/2.8L IS II USM Telephoto Zoom Lens
Slik 700DX Tripod
Bogen 3030 3-way pan/tilt head

Be Sociable, Share!

9 thoughts on “How clean can you get ISO 102,400?

  1. - Gregory Allen Deese

  2. I’ll have to try this. It’s counter-intuitive to me because usually when i’m working on HDR images noise is increased during the process of combining photos not decreased. perhaps there is something else going on or you are referring to a diffferent method of combination than photomatix.

    • Hi Jason,
      Definitely don’t try this using photomatix, it’ll give you more noise! Take your over exposed image(s) and process them a second time but this time decrease the exposure to the next image down. For example, 3 images spaced 2 stops apart, A (under exposed), B (proper exposure), C (over exposed). Process B and C but drop their exposure in development to match that of A and B. You will then combine A + B-dropped using layers. What you want is just the shadow darks areas from B-dropped to replace the shadows dark areas of A. That’s the simple quick explanation, I need to write up a longer post explaining in more detail and everything that you can do to decrease the noise. Still haven’t had time to fully test the new method I’m thinking of trying, it’s many more steps but should give better results.

    • Ok, let’s take a bracket of 3 shots as an example. You have three images that range in exposure value: -, =, and +. What you are going to do is combine the current image with the next image to the right that’s exposed more. So the – image will be combined with the = image. The = image will be combined with the + shot. The + shot we have nothing to combine with it. As an example we’ll take the – and = shot that we want to combine. If the bracket was shot 2 stops apart, you’d process the – image normally. You’d also process an extra version of the = image but 2 stops underexposed. Load the two images as layers in photoshop, the – image being the bottom image. Put a layer mask on the = image that was processed 2 stops under, the layer mask being a grey scale version of the layer. On the layer mask you’ll do a threshold, this way we can keep the dark areas of the =_2_stops_underexposed layer but the bright areas we’ll see through that layer to the underlying – layer. The threshold value you can play around with to see what works best for your image. Once that’s done do a slight blur on the layer mask so the transitions are smooth and you are done.

  3. That is some pretty amazing result. Even if I think I understand the process you went through, I would love a detailed tutorial in a next article explaining step by step what you did. I will try on my own though and hope I’ll get some similar results. Great job!

  4. The Triple Bypass Photography Link List » Latest News Update | Latest News Update

  5. The Triple Bypass Photography Link List

Leave a Reply