A reader suggested that a High-Res Olympus E-M5 Mark II image used in the previous post looked sharper than the equivalent Sony a6000 image, contradicting the relative MTF50 measurements, perhaps showing ‘the limitations of MTF50 as a methodology’. That would be surprising because MTF50 normally correlates quite well with perceived sharpness, so I decided to check this particular case out.
‘Who are you going to believe, me or your lying eyes’?
MTF measurements are by no means perfect but when used judiciously they have proven over half a century to be quantitative, objective and relatively accurate. ‘Relatively’ only because they rely on the images tested being captured with good technique, which is not easy to do given the large number of variables involved. As with all science, part of the job is to discard obvious outliers.
So let’s see if in this case, as our reader maintains, “it’s clear that the real-world resolution advantage lies with the E-M5” II despite MTF50 readings off the raw data showing the a6000 ahead by 15% or so. Recall that a 5% difference in MTF50 may be hard to spot by eye in two images side by side but 10% or more is definitely noticeable.
I took the ISO200 ‘Daylight’ raw files for both the a6000 and the E-M5II in 64MP High Res Shot mode, in my opinion the more directly comparable pair of images of the relative DPR New Studio Scene series.
Next I opened both raw files using standard procedure and dcraw -d -4 -T -w**. This produces a grayscale image from the unprocessed raw data with each color channel normalized to the same peak intensity. In neutral portions of the resulting image we are therefore looking at the actual physical capabilities of the hardware, before any processing – exactly what I am interested in, because otherwise we know that sharpness may be changed arbitrarily/inadvertently/unknowingly through in-camera or in-converter processing .
We also know that some raw converters, including unfortunately the one used by DPR, apply different types and amounts of processing depending on the camera. Even with all sliders set at zero – noise reduction, sharpening, lateral chromatic aberrations, distortion corrections may be applied in different amounts to different cameras during conversion, distorting the actual spatial resolution information captured in the raw file.
Having introduced many non-linear arbitrary variables of arbitrary strength it becomes therefore difficult to fairly compare the capabilities of the hardware of different systems when qualitatively looking at their output, even at the moderate precision that we are after. On the other hand raw data contains the real, objective unadultereted deal. That’s why I measure the MTF in the unprocessed raw data.
The idea is to capture the best spatial resolution information possible in the raw data comfortable in the knowledge that with better objective ‘sharpness’ information to start with we are going to achieve better final results after subjective processing.
In order to properly compare two images they need to be viewed at the same final display size (or customarily picture height). So next the a6000 image (4024 pixels on the short side) was upscaled to the size of the E-M5II (6938) using Photoshop bicubic.
This is the portion of the raw captures that allegedly shows the E-M5II to be ‘sharper’ than the a6000 contrary to MTF50 data:
The a6000 image looks a little aliased, as expected, but it also looks sharper as predicted by the MTF50 readings outlined in the previous post. Click on the image to view it at 100%, ensuring no zooming by your browser (ctrl-zero works for most).
Here is the center of the image for a different take:
Once again the E-M5 mk II’s High-Res image produced by dcraw looks softer, as predicted by the relative MTF50 readings. So I believe that the usefulness of MTF50 as an unbiased arbiter of image ‘sharpness’ has been reaffirmed, its reputation untarnished.
I am satisfied therefore that even in this case MTF50 measurements correlate well with perceived sharpness, confirming the value of the Modulation Transfer Function approach as a means to objectively quantifying linear spatial resolution measurements.
** As mentioned earlier, if dcraw 9.22 did not do a good job of unwrapping the raw files the absolute comparisons in this and the previous post should be disregarded. [Edit: Since the latest version of RawDigger 1.1.3, which supports EM5II’s HR shot mode, produces similar results I am starting to feel more comfortable that we are good to go]