Noiseless images with the OM-D - well, almost




ISO200|50mm f1.4|1/2500s|f4.0

Blue sky with branches - should be smooth, yes?

OK folks, I have been researching this issue/feature a couple of weeks now and although I am not completely done with the research yet, the new firmware update has made me publish this right now.

So you are an enthousiast on a budget. Have shot film, know the trade, sensitive to quality. Researched all the options and decided that a) mirrorless useful at last and b) considering the lenses and all, Olympus is the way to go. One reason for that is the very nice dynamic range, the One Thing we care about nowadays (since the "noise issue" seems to be sorted by the transition to CMOS, anyway). Purchased the gear, love the features, small size, etc. You shoot your considered shots and than - bang, you discover that there is no such thing as a noiseless iso. Bummer.

Now that firmware update 2.0 is out, you are very happy: now we DO have a low iso (reported as iso100). You upgrade, fire up Lightroom and - argh. It is better, but still not noiseless. Bummer - again. The crop of the shot above (see below) illustrates the point (note that the clouds were moving and thus one image appears to be brighter than the other, but the overall luminance of the two images are roughly the same - as the coming images, these are illustrations of a point you can try for yourself).


ISO200|50mm f1.4|1/2500s|f4.0

Compare the base iso200 to the new iso100 by Olympus with the (hover mouse / touch image)
(100% crop from the top right area of the first image (RPP conversion without "blur chroma" or "highlight recovery", saved as 90% jpg.).

So what do you do? I answer in two parts. The first is the practical ("cookbook") and the second is the theoretical.

Cookbook: how to get "noiseless" images with the OM-D in five easy steps

  1. Make sure you are shooting RAW. Ask yourself: is this image going to be (really) high dynamic range? If the answer is "no" go forward.
  2. Meter the scene and look at the historgram/flashing indicator/preview image. Are you happy with your highlights? Make sure you leave some room for the highlights, that is do not "expose to the right" aggressively.
  3. Based on your readings in ii), apply two EV exposure bias (you have to be in either A/S/M mode for this, do some thinking in the latter case). Do _not_ worry about historgram, etc.
  4. Now you'll have to have the shutter open for four times longer than you originally thought. Ask yourself: is it OK to do so? (No moving subjects, etc. OR enough light at your disposal) If the answer is "yes", expose. (Forget the ugly image you see in the EVF.)
  5. Fire up your RAW converter. If you are using an honest converter, such as RPP, you are good to go without extra adjustments (in fact, this is the first time you'll se a well exposed image). If you are using Lightroom 4, apply -2EV adjustment. If you are using another software, get to know it and figure out how much does it cheat. Apply the negative as an EV adjustment. Enjoy your nice noiseless image!

You can compare the results to the new Olympus iso100 sensitivity below. Looks like an improvement to me! No wonder: iso50 is nicer than iso100 and more light is better than less! Try this in various circumstances. Today, I have tried this for headshots in strobe light and there was a clear improvement in shadow areas. These are subtle changes, but if you print 1:1 on paper, they'll definitely show. Clearly, they'll show even more if you open up some shadows in post.


ISO200|50mm f1.4|1/640s|f4.0

100% crop from the top right area of the first image. Compare the new iso100 by Olympus with the overexposed and adjusted base iso200 (hover mouse / thouch image)
(100% crop from the top right area of the first image (RSP conversion without "blur chroma" or "highlight recovery", saved as 90% jpg).

Things are more difficult in a high dynamic range situation. My example shows a case where (assume) I want to show texture of a chair shot against light and keep the background detail at the same time (a rather challenging task I must say - the right way to deal with it is of course using light - most of the time).


ISO200|50mm f1.4|1/125s|f2.0

Dark chair against light - almost completely dark without the curve adjustment below

There is no easy win here, but we can still achieve some progress by noting that we are planning to open up the shadows. The sharpening and noise reduction facilities, along with the ability to use flexible curves in Lightroom makes it a breeze to do. Even though the dynamic range is high here, I could achieve less noise by overexposing two full stops, then pulling the whole scene and restoring the shadows using curves. This time, there is noise reduction and sharpening in both images. (Forgive me for the CA - LR could have taken care of that, but cutting out the same part was not easy and I do not want to repeat.)


ISO200|50mm f1.4|1/125s|f2.0

Compare the base iso200 by Olympus with the overexposed and adjusted base iso200 (hover mouse / touch image)
(100% crop, LR conversion with sharpening and noise reduction applied).

You might want to ask if there is a point to this (first) cumbersome procedure when there is the all new iso100? Yes! Although the results are most striking at iso200 (now iso50), you can do this at iso1600 or iso3200 or even above! Of course, you will have an effective iso400 and iso800, etc. respectively, but still, you do better than before. The second example shows that even if you are planning to do what the camera would do (opening up the shadows), modern RAW processing technology lets you to do this better than the camera.

I know this might sound a bit disappointing. Before your Olympus ego starts to get sad, note that other manufacturers do this too. They may make different decisions and working with larger or smaller sensors, they might have more or less room to do so. I think that considering the general use of the camera, the decision Olympus made is wise.

(Note: I really _love_ new new small focus points! ;) )

The theory

Note that I do not _really_ know what I am talking about here. There are others who do - go and read them. A good starting point is the online photographer's take on iso. In fact, I recommend you reading that piece and coming back here thereafter.

Right, now you know that iso is a manufacturer's decision. You probably also know that there is technological progress, but no free lunch. Based on current technology, there will always be a trade-off between dynamic range and noise. If you buy a small sensor camera, you will have tighter constraints than the guy buying a 35mm, not to talk about medium format. Following their target group (in the case of Olympus and the OM-D: serious amateurs on a budget), manufacturers make their decision on this trade-off. People at Olympus have decided that they trade noise for dynamic range (in the final jpg, that is - the sensor's capabilities are fixed). They strongly underexpose and apply a strong curve adjustment to "open up" shadows, or more appropriately: everything except for highlights. As the sensor is very nice, this works well.

(Note that this is no different from the method we used to use some five years ago, when this was _not_ general practice. That time, we exposed aggressively to the right to protect highlights and opened the shadows in the RAW processor. Now the camera does this by default, like it or not.)

(Also note that this practice is not very different from the rating of films done in the film days. Serious amateurs used to fine-tune their options by rating film, exploiting trade-offs similar to the one shown here.)

As always, there is a price to pay for decisions and in this case it is apparent noise even at blazing sun. What can you do? Well, I have already told you. To get optimal quality, now you have to do the opposite of you did a couple of years before. Decide if you really need that high dynamic range and if not, trade it for less noise. You can do it!

Remark: I am even less sure about this, but I think that all this has to do with sensor technology too. CMOS sensors are noisy by default and the trick is to use exposure-time noise reduction using a sample of real-time background noise. Because of the way this works, such noise reduction is way more effective than the one done in post-production. This implies that the trick camera-makers are doing here is relatively expensive in terms of noise, as it is effectively done in post-production. By doing the overexposure trick, you limit dynamic range consciously, but harness the power of on-chip noise-reduction much better.