Keeping it Sharp
From SarahWiki
I've been into the photography thing for more or less as long as I can remember, with (so I thought, until recently) plenty of experience with moderately decent equipment, so I didn't expect difficulty in getting good results when using my new Megavision E4 monochrome back. Well, my results were mixed, and it turned out to be entirely my own fault, so I'm writing this essay in the hope that what I learned in the process of tracking down the problems might be of use to someone else.
Back in the good old days with film, if the image was OK in the viewfinder, it tended to also be perfectly OK on film. Any slight inaccuracy was easily lost in the grain, even with slow film. Modern digital sensors, assuming reasonable amounts of light, have essentially no grain whatsoever, and considerably finer resolution even than the very best film. This is actually a good news/bad news thing. Whilst it makes it possible to achieve far better image quality than film, any minor defects either in technique or equipment tend to be glaringly apparent, and the better the sensor, the greater this effect tends to be.
Contents |
Pixel peeping
Much is said about the evils of pixel peeping, i.e. the tendency of people to look at images at a 1:1 pixel resolution in order to check for problems. Granted, an image that looks unusable 1:1 is often still printable at small or moderate size and also quite usable for web purposes. Nevertheless, if you've dropped a huge amount of money on state-of-the-art kit, if you're off by a 2:1 factor (which is easy to do), with a circle of confusion covering two pixels rather than one, you might as well have an image sensor with quarter of the resolution. A 16 megapixel sensor in such circumstances is effectively delivering about 4 megapixels. In my opinion, this is a waste of money and effort. I went for really high end kit so I could get outstanding results, so I try always to go for as-good-as-possible, rather than good-enough.
There is an analogy here to my past life in audio engineering. I once worked on designing pro audio equipment, where there are (as in photography) very much two schools of thought: the 'scientific' approach, where design centres around frequency response, phase response and distortion measurements, and the 'arty' school where people design by ear and spend hours hand-selecting components and tweaking the circuit to get the best sound. Actually, from experience, both approaches can be shown to work. The arty approach works because someone with good ears took the time to tweak the circuit so that it sounded good. The scientific approach works because, well, if you amplify an audio signal and your amplifier has huge excess bandwidth, flat phase response and near-unmeasurable distortion, there is basically no way for it not to sound good -- bad sound isn't magical, it always has to result from some kind of deviation from the ideal waveform, so if no such deviation exists, the sound doesn't have any way to be bad. Whilst some listeners might prefer the sound of 'arty' circuits (in particular, even harmonic distortion sounds pleasant to many people), scientifically designed circuits just work, and are certainly far easier to design and manufacture repeatably.
In photography, I find that if I pixel-peep and discover that my image is blurred, if I fix that so that the image is sharp down to the pixel level, the prints are pretty much always really good. Otherwise, my prints are sometimes good, but not reliably so. As with audio, it's really a repeatability thing. These days, I now make a habit of checking focus by using the Megavision software to zoom in to critical bits of the image to at least 400% -- if lines and small features are 1 pixel wide, I'm there. If not, I try to fix it somehow. I'll come on to the how in the next section.
Focus pocus
The most critical thing in achieving sharpness, perhaps unsurprisingly, is focusing. These days, I suppose it is possible that many people may never have actually focused a camera by hand, given the great rise of autofocus technology from the early 1990s onwards. Nevertheless, even now, all view cameras and most medium format cameras have manual focus only, and even with AF, it is often the case that the camera will focus on the wrong part of the image or just not quite get it right, resulting in the need to switch to manual anyway. Personally, I actually prefer manual focus, I never really quite got to like AF very much because it tends to make me work a bit too fast, not really quite consider what I'm seeing properly and as a consequence make too many stupid errors.
Focusing for film is easy enough. If it's OK in the viewfinder, it'll be fine on film too, and you don't really need to worry about it, particularly if you have the aperture stopped down a bit. This doesn't really apply to high resolution digital sensors, unfortunately. Slight focus errors, such that would not be apparent at all on film, are often very visible indeed. I find that when focusing on ground glass, I need to exert quite a bit more effort to get the focus point just right, or I end up being disappointed with the result. With my Bronica ETRS, I find having a focusing screen with a split image rangefinder prism extremely useful -- if I'm using a prism finder, it's still a bit tricky even then. My preference more recently is to use a waist level finder, which incorporates a magnifying lens that gives a much bigger and brighter image than the prism finders, which makes focusing considerably easier and more accurate.
Depth of field, or rather the lack of it
Depth of field calculations are traditionally made assuming quite a large circle of confusion, something that fits well with typical film grain, but that is rather too large when applied to digital sensors. I should fact-check this, but I seem to remember that the circle of confusion used in calculations is typically about 30 microns in diameter. With a 9 micron pitch sensor, that's enough for a 3x3 pixel grid to fit inside it, which would make a 9 megapixel sensor work like a 1 megapixel sensor, at least in theory. For colour imaging, it's not really quite as bad as this looks -- colour CCDs use what's known as a Bayer matrix, where each 2x2 block of pixels contains two green pixels, normally aligned diagonally, one red pixel and one blue pixel. On its own, this works, but isn't really fine enough resolution to avoid bad colour aliasing problems, often known as the christmas tree effect, where an image artefact smaller than the 2x2 resolution covers only part of the Bayer matrix, thereby resulting in significantly incorrect colour. As a result, most digital colour sensors incorporate a low pass filter, basically an optical component designed to diffuse the image by just enough to avoid aliasing problems from the Bayer matrix. In practice, this usually involves a roughly 2:1 reduction in fine resolution in each axis, or about a 4:1 reduction in effective pixel count. A square 16 megapixel colour sensor therefore has an effective resolution of about 2000x2000 full colour pixels, but interpolation is normally used to deliver an image with 16 million pixels. True monochrome sensors are extremely rare, though they do exist -- my Megavision E4 monochrome sensor has a genuine 4091x4091 resolution with no interpolation or low pass filter, and all of my Bronica lenses seem able to keep up, given careful use. That's the good news. The bad news is that, from the point of view of depth of field, I have to work with a circle of confusion half the diameter and quarter of the area that I'd be able to get away with colour, or maybe a third of the diameter and a ninth of the area that would be possible with film. This makes for a really quite significant difference in the way I work. Where previously I'd crack the focus off a bit using the depth of field scale on the lens, I've since learned the hard way that this really does not work, so if I want something sharp, I have to focus on it, and that's that. I am very glad to own a shift/tilt lens, actually -- for landscape work, it is very difficult to get a large area of ground in sharp focus with conventional lenses, but it tends to be much easier if you can add a little bit of tilt. As with focusing though, getting the right amount of tilt can be quite difficult, so I now tend to make an initial attempt by eye through the viewfinder, then check focus across the whole image by scrolling around and looking at the image at at least 400% magnification.
Aperture
Depth of field is traditionally controlled by setting the aperture -- higher f numbers for more depth of field, lower f numbers for less. With film, most decent lenses are usable across most if not all of the f-stop range. Wide open, some lenses looked a bit soft, but in general this wasn't generally a problem. However, getting a sharp image down at the 9 micron pixel dimension is a taller order. Wide open, most lenses are too soft, so it's necessary to stop down by at least 1 or 2 stops. I found that my Bronica primes were actually pretty good wide open, particularly the 40mm and 75mm. The 100mm macro, 150mm and particularly the 250mm do need a bit of stopping down -- the 250 really isn't that nice below f/16. The 45-90mm zoom is a little soft at its widest aperture setting, so it likes at least f/8 at 45mm or f/11 at 90mm, but like the primes, it's pin sharp down to the pixel level. So much for zooms not being able to replicate the sharpness of primes! The shift/tilt lens definitely doesn't like to be wide open very much, but that's not terribly surprising since it's an adaptation of the Schneider Super Angulon large format lens, which was never intended to be used wide open anyway. The reason for this effect, in all cases, is that machining a large expanse of glass extremely accurately is very difficult, so small aberrations in the surface contribute to a general reduction in sharpness. Smaller apertures counteract this by effectively reducing the amount of (mostly internal rather than external) lens surface that affects the image. Good quality 35mm and medium format lenses need very little, if any, stopping down -- my tests indicate that the Bronica glass is probably going to be perfectly OK in most cases with a colour sensor, and almost certainly fine wide open with film. The monochrome back typically needs to be stopped down 1 or 2 stops, but that's certainly not really an operational problem.
On the small aperture side, a second issue comes into play -- diffraction. This is a tougher problem, which is not really amenable to solution. Basically, light wants to bend around the side of any hard edge that it comes across. If you shine light across a knife edge, some light will go straight on, and some will go 'around the corner', so-to-speak. Similarly, if you shine light through a narrow slit, you typically get a fuzzy line, not a sharply defined line, at the other side. This is an example of light deciding to behave in a wavelike, rather than particle-like way. If you send light through a small hole, as with a pinhole camera, overall resoution is limited by diffraction effects, such that making the hole smaller, beyond a certain size, fails to make the image any sharper. Unfortunately, a lens aperture iris behaves just like a small hole as far as light is concerned, because that's actually exactly what it is. As a consequence, diffraction effects start to predominate when lenses are stopped down to small apertures, with the effect that the smaller the aperture becomes, the more the image sharpness is lost. This means that, in most cases, with the monochrome sensor, I start to lose resolution at about f/22, and this gets really quite noticeable as you stop down further. As a consequence, just stopping down in order to increase depth of field is often not really an option, because the diffraction losses look even worse than inadequate depth of field.
Camera shake
Since it's necessary to be so much fussier about focus and depth of field, it should not really be much of a surprise that camera shake is also a significant issue.
In my film days, I spent quite a while too poor to afford a tripod, so I got pretty good at hand-holding both 35mm and medium format cameras. Perhaps unsurprisingly, I've found that with the Megavision back, I need a good solid tripod at the very least in order to stand a reasonable chance of sharp results. I'm currently using a Giottos carbon fibre tripod with a 3-axis head, which is pretty solid by any reasonable standards, though I find that I get about 1 pixel of vertical blur if I use the shutter release on either a motor winder or the speed grip. The shutter release on the camera body is slightly better, but still not ideal. I therefore now pretty much always use a cable release to fire the shutter, after giving the rig a second or two to settle down after last touching it. This seems to be enough, which is just as well -- the ETRS I use doesn't have mirror lockup, and gives off a hefty clunk/bang when you fire the shutter. The noise is nearly all mirror, of course -- since the shutters are all buried inside the lenses, the actual shutter is extremely quiet.
Conclusions
I feel a bit like I'm learning photography from scratch again. I'm finding nearly all of my assumptions being challenged, and many of them overturned. It's a steep learning curve, but my methods seem to be improving, which is giving a direct return in the quality of my images. In summary:
- Use good glass
- Stop down enough to hit the sensor's resolution limit, but not enough to cause diffraction losses.
- Use a really solid tripod
- Use a cable release
This seems to work, but it's quite a departure from anything I've previously needed to do. The results are truly worth it, though -- the impact of this kind of resolution in a large print can be truly stunning.
