Conventional wisdom states that the bigger the sensor, the better. The bigger the pixels, the better. All things equal, that’s true; however, 10-micron pixels would mean very low resolution compacts, and medium format digital doesn’t sell in sufficient volumes to justify the same sort of R&D spend that consumer or even midrange pro gear would get. I admit I’d always been curious to see just how much the technological improvements from generation to generation offset pixel pitch etc.; some time ago, I did a comparison of the Leica S2 against the then-new Nikon D800E. Today, we go one step further to see exactly what kind of gap exists between the various grades of equipment. Spoiler: it’s not as wide as you might imagine in some areas.
Important note: We’re not strictly comparing like with like (the CFV’s sensor architecture is many generations behind the D800E), and although the test is as scientific as I can make it – tripod use, mirror lockup etc., the priority is always to make it as realistic a comparison as possible. This means the cameras were shot with the settings I’d usually choose, and some processing is necessary, because I’d never use untouched files; however, where such processing was done, it was done as fairly as possible for all cameras – I’ll explain more in detail as we go along. Please look at the linked images to go along with the commentary.
Relative sensor sizes. If the Hasselblad had the same pixel pitch as the Canon, it’d have a mind-boggling 1.07 billion pixels.
In the mix today are five cameras with sensor sizes ranging from jumbo (49.1x37mm) to minuscule (4.8×3.6mm effective) and spanning several generations of technology. The oldest – mid-2009 – is also the largest; the Hasselblad CFV-39 medium format digital back uses a Kodak CCD with 6.8 micron pixels. The sensor has no AA filter, but also no microlenses; this means purple fringing is kept to a minimum, but light-gathering efficiency suffers as a result. The CFV consistently required about a stop more light than the other cameras to achieve the same histogram exposure. The Nikon D800E has perhaps the best full-frame sensor going at the moment; it’s a CMOS known for incredible dynamic range and a very low noise floor. The OM-D’s 15MP Sony CMOS’ strength is color reproduction, though I suspect that’s as much due to Olympus’ algorithms as the sensor; curiously, it has nearly the same pixel pitch as the D800E – though in practice, tends to be about a stop noisier at the pixel level. The Fuji uses a proprietary EXR-CMOS sensor (not the X-Trans array) – whose RAW output ACR still can’t seem to get right, so I shot JPEG for this test – as befits how I’d normally use this camera. Finally, we have the slightly oddball Canon: though the sensor is a 1/2.5″, 16MP type, the design of the camera’s optics – part extending barrel, part internal lateral movement with a 90deg light path change via a prism to keep the camera 19mm thick and still offer a 28-330mm optical zoom – meant that it only uses the central 10MP portion, for a very, very small effective sensor size. If you’re wondering why these particular cameras, well, it’s what I have to hand at the moment. I’d have liked to add the RX100 and D7100 for two more intermediate sensor sizes, but I currently don’t own either.
I’m looking at a few things: firstly, dynamic range; secondly, resolving power; and finally, the slightly less exact science of tonal reproduction, color accuracy and tonality on monochrome conversion. The highest possible shot discipline was used for each camera – optimum apertures (f16 for the Hasselblad, f8 for the Nikon, f5.6 for the OM-D and XF1, and whatever the Canon chose – it doesn’t have aperture priority); mirror lockup/ timer where applicable and finally lockdown on my usual massive Gitzo 5-series studio tripod White balance was matched by using the eyedropper tool on the same grey area.
What looks like a fairly flat scene is actually a torture test: the D800e’s spot meter made the difference between the brightest and darkest portions to be around 9 stops. It’s backlit and fairly harsh tropical sun.
Overall comparison with histograms; click here for a larger version.
Exposure was adjusted via on-board histogram to try and make the highlights just clip directly over the sun; the smaller sensor cameras had problems with this as there appeared to be some leakage/ blooming into adjacent areas. No recovery was used in ACR. The histograms tell more of the story than the overview image; it’s clear that the larger sensors are doing the best here, with the CFV having dynamic range to spare on both sides of the histogram, and the D800E matching it in the shadows. Surprisingly, the OM-D is not far behind the D800E, despite yielding a slightly brighter image overall. Our two smaller-sensored cameras did surprisingly well considering that dynamic range correlates almost directly with electron well size and physical pixel pitch. The histograms don’t tell the whole story, though – the way the cameras render the clouds is telling; there’s little highlight separation left in all the cameras except the CFV and curiously, the XF1.
Noise floor. Click here for a larger version.
These results are more like we’d expect: generally, the larger the pixels, the lower the shadow read noise. Older technology – the CFV’s CCD – offsets that advantage somewhat. Unquestionably, the D800E has the cleanest shadows and also the most accurate color; the CFV is showing traces of chroma noise and a slightly cyanotic cast. The OM-D is clearly noisier than the D800E, but still maintains good color; the compacts, on the other hand, are a mushy mess with almost all detail lost to noise and then poorly applied noise reduction algorithms. A clear win for CMOS and newer technology here.
Resolving power and acuity
Test scene, viewed at identical magnification for all cameras. If pixels were equal and the Hasselblad image was printed to 60×45″, which it can do easily, the Canon should be able to make 30×22″…in reality, I think 22×16″ is about the limit for the Canon.
If all pixels were of equal quality, and you printed each file at the same DPI level, this would be the difference in maximum size – you need to go to the CFV from the Canon to double your print in height and width. No surprises there. However, there’s a catch: due to AA filters, lens quality, pixel size etc – pixels are nowhere near equal. Lenses used – in an attempt to match FOV as closely as possible – the Zeiss CF T* 120/4 Makro-Planar on the Hasselblad; the Nikon 85/2.8 PCE on the D800E, and the Olympus 45/1.8 on the OM-D. No choice of lens on the two compacts, obviously.
Click here for 100% crops
This crop makes it pretty clear that at a pixel level, the CFV wipes the floor with the other cameras (note: no sharpening was applied post-capture; the XF1 and Canon both have default JPEG sharpening that can’t be turned off completely). Just look at the trees and the texture in the side of the building. The D800E comes reasonably close, but doesn’t quite make it – there’s a bit of clarity missing in comparison, as though there’s a very thin veil over the image. Look at the side of the building. It’s not the lens or camera, though; this is consistent with the other 35,000+ images I’ve shot with the camera, and as good as anything I’ve ever seen out of it. The OM-D comes surprisingly close to the D800E at the pixel level; actually, not that much of a surprise since their pixels are reasonably close in size. Our two compacts are lagging despite rather enthusiastic automatic JPEG sharpening. Once again: it’s clear that pixel size still matters, even when we remove the anti-aliasing filter from the equation. The scientific explanation for this is that perceived sharpness is dependent on edge acuity; and edge acuity is dependent on signal-to-noise ratio. The bigger the pixels, the better the signal-to-noise ratio.
Click here for 100% crops
The difference is even more stark in the second crop when there are clear high-frequency, fine-detail structures – just look at the text…
Here we start getting into difficult territory: firstly, it’s necessary to explain the processing performed. This is meant to represent a realistic scenario: I wouldn’t use an image without optimizing it first, but even after Photoshop, there are some tonal properties that cannot be added if they weren’t originally present – such as microcontrast and the character of tonal response in particular zones. What I did was open all images in ACR; set white balance off the grey card you see peeking in the lower right corner, and maximize the tonal range used by adjusting exposure and auto-levels to ensure the full tonal output range was being used. Dynamic range of all cameras was sufficient not to clip any highlights or shadows in this scene. It’s important to note that tonal and dynamic range presentation is also impacted by the bit depth of the camera: the more bits, the more output ‘buckets’ you have to allocate tones. The Hasselblad is a 16 bit (though using 15) camera; the OM-D and D800E are 14-bit, and the Fuji and Canon both limited to 8 bits due to JPEG output.
Click here for a larger version.
Look carefully at the petals: which appear to have the most texture to you? This is microcontrast; I suppose it’s analogous to the ability to display subtle tonal gradations at a high frequency. It’s partly a lens property (in that it requires resolution of fine detail structures) and partially a sensor one – even if the lens can resolve the detail, the sensor must have the available tonal range to record subtle luminance/ color changes. These changes are what we perceive as texture or dimensionality. In my eyes, the CFV wins here by a relatively small margin. The D800E and OM-D are tied for second, with the Fuji surprisingly close; its sensor seems to protect the highlights at the expense of the shadows – even the Canon has more detail in the leaves. The Canon’s output just looks waxy – it’s a combination of a lack of resolving power AND a lack of tonal range, but very typical for consumer-grade output. Interestingly, none of the cameras quite get the color right – but if I had to pick, I’d give it to the Olympus by a hair; the CFV is too cyanotic, the D800E too yellow, the Fuji too dark, and the Canon seems to be struggling with gamut – not surprising as its output is SRGB-only. Note that none of the these cameras are running my custom profiles; they were all converted with ACR defaults.
Second color sample. Click here for larger. Nothing much changes in terms of observations, but they are a lot closer here – the colors being closer to core gamut for all cameras.
Something almost all cameras struggle with are the reds: this is typically due to near-IR spectrum pollution of the red channel, which then causes the camera to record more than what’s in the visible spectrum, resulting in oversaturation and clipping. Modern cameras tend to be much less prone to this, largely thanks to improvements in the filter pack in front of the sensor.
Red saturation. Click here for larger.
All of the cameras do a pretty good job, with the exception of the Fuji. It’s clipped to full 255/0/0 in several areas, and little detail remains in others. (The vase has a pebbled surface texture best represented by the D800E’s swatch.) Here, the CFV shows the age of its technology: tonal gradation and color aren’t as accurate as the D800E, and it doesn’t have as much shadow detail in that channel either, even though it has larger pixels. The OM-D is quite close to the D800E, and the Canon is doing a surprisingly good job considering how small the pixels are – it might well have the strongest IR cut filter of the bunch.
Black and white conversion potential
Click here for larger.
Black and white tonality is something I talk about quite a lot, but perhaps one of the most difficult things to define accurately. I suppose it correlates best with three other properties – dynamic range, native tonal response, and microcontrast. You need wide dynamic range for a natural-looking image with smooth shadow and highlight transitions; the native tonal response of the sensor must be biased in such a way as to create highlight and shadow ‘ramps’ to help smooth the transition to pure white or pure black; especially important since you no longer have the distraction of color and partial saturation to temper the impression of a harsh cutoff. The best place to see this is the petals themselves: do they look curved, textured and three-dimensional, or ‘folded’? Microcontrast is the ability to reproduce low contrast, high-frequency detail structures – like the wall or petal texture.
In my mind, there’s no question the CFV wins here; CCDs have a bit of an advantage over CMOS sensors as their tonal map is definitely nonlinear; CMOS sensors tend to be. That said, there’s really not a lot in it, and if you have a poor monitor calibration or a bad print, you may well see zero difference between the CFV and the Canon! The Nikon remains a hair better than the Olympus because of its greater dynamic range; some of the petals in the Olympus swatch are starting to look flat. Surprisingly, the XF1 appears to be holding its own with the Olympus; whatever Fuji have done to their sensor, it appears to be very highlight-biased; there is very little shadow information whatsoever. Finally, we have the Canon: it gives you black and white, literally. There is no room for grey.
I’m surprised not that the largest sensor with the biggest pixels wins in most categories, but that the results are so close at all; small sensors have really become quite good. (In fact, I suspect that if I’d had an RX100 handy, it’d probably be quite close to the OM-D despite having pixels similar in size to the Fuji.) The reality is that at moderate print sizes – removing resolution from the equation – differences in exposure and processing choices from one photographer to the next can well erase or even reverse the differences we see here. What this comparison does not take into account is that the compacts and the CFV must be shot at base ISO to achieve the best results, with considerable compromises above this; the D800E and OM-D will happily go to ISO 1600 without much penalty – making them significantly more versatile cameras. They also have more malleable RAW files – the D800E’s files especially seem to be very resilient and forgiving of exposure errors. The Hasselblad, by comparison, is perhaps the most unforgiving camera I’ve ever shot with – it offers very, very little latitude for correction.
The only conclusion I can come to is that under ideal circumstances, most users may not see that much difference in climbing the diminishing returns curve; certainly not enough to justify the added hassle and expense. In real terms, the Canon can be had for $130 or so; the Fuji $400; the OM-D with a couple of good primes, $1500+/-; the D800E body is $3,300, and probably another $1000 for lenses, and don’t even ask about the Hasselblad – the current CFV-50 (50MP) model is a whopping $17,000 – and that’s just for the back, with no camera and no lenses. Is it six times better than the D800E? No. Is the Fuji 2.5x as good as the Canon? Possibly. The OM-D is still enough of an improvement over the two compacts, and surprisingly close to the D800E on a pixel level. But unless you have physically very large output requirements – it’s tough to recommend upping the investment. Don’t forget, it doesn’t stop at the direct hardware: you need to consider adequate tripod support, computing power, storage, carrying, etc…it never ends.
That said, I find the Hasselblad perhaps the most rewarding of the five: when you get it right, no correction is necessary. The above sunset was almost straight out of camera; it’s the first time I’ve been able to correctly capture the tones and overall feeling at the time. (JPEG/ web compression reduces the overall impact significantly compared to the 16-bit TIFF on my calibrated monitor, though.) Interestingly, no other camera I’ve used has been able to do this – including the D800E. I can only chalk it down to all of those little differences adding up. MT
The equipment used in this review is available via the following links: Hasselblad CFV-50 (B&H); Nikon D800E (B&H, Amazon); Olympus OM-D (B&H, Amazon); Fuji XF1 (B&H, Amazon); Canon 520 HS (B&H, Amazon).
Visit the Teaching Store to up your photographic game – including workshop and Photoshop Workflow videos and the customized Email School of Photography; or go mobile with the Photography Compendium for iPad. You can also get your gear from B&H and Amazon. Prices are the same as normal, however a small portion of your purchase value is referred back to me. Thanks!
Images and content copyright Ming Thein | mingthein.com 2012 onwards. All rights reserved