In assessing the image quality I’m seeing recently from the Olympus OM-D E-M5 under controlled conditions (ditto for the Sigma DP1 Merrill), I have to say that quality can be very high with proper exposure, on par with most DSLRs.
It has become clear that many of my field shots the E-M5 deemed to have a “correct” exposure could have accepted a full stop, perhaps as much as two stops more exposure in some cases. I would say that 80% of the images could have accepted at least a full stop more exposure, the exceptions being high contrast images with bright highlights and dark shadows. A full stop means a noise difference of 1.4X, two stops means noise is cut in half.
The foregoing is true even though I almost always shoot manually using ETTR techniques and add exposure over and above the camera recommendation.
The problem is that the Olympus E-M5 is so misleading in its histogram that a wholesale reevaluation of technique with the E-M5 should be on the menu for any E-M5 shooter: the E-M5 tends to badly underexpose, even if the image looks reasonably exposed; the sensor can accept a lot more exposure, and then deliver much superior image quality (shooting RAW is mandatory of course).
This underexposure behavior does not seem to be peculiar to the Olympus E-M5.
I have observed the behavior extensively in the field (not just the Dolls scene).
Why is the exposure so poor? Perhaps because camera vendors tune metering to JPEG shooters, polluting the menus and controls with JPEG-related dreck, including the all important histogram. With conservative exposure, a JPEG avoids blowing out. But as the pair of images below shows, the best exposure for RAW will badly blow-out the JPEG. Hence sub-optimal exposure for RAW, with up to twice the noise (or more) than need be (mitigated by JPEG, which blurs away fine details like noise!).
Accordingly, my future camera assessments from now on will include an assessment of how much exposure the sensor can take, versus what the camera histogram claims is full exposure.
The really good news is that by maximizing “signal to noise”, very high image quality can be obtained from small sensor cameras. Of course it is not sensor size per se— a future 60 megapixel DSLR with tiny relatively noisy photosites would be in the same boat.