Introduction

Ever since I posted my review of the Canon EOS-5d, I've gotten quite a few emailed questions from people who are considering jumping onto the 'digital bandwagon'.  Far and away, the question most asked is "What's the largest print you can make from an exposure from the 5d and get a print of acceptable quality?"

As it turns out, this is a simple question without a simple answer.  Part of the problem is that print quality is not a simple scalar value; it's composed of a number of properties which combine in a non-linear way to produce the final 'quality' of the print.  The other part of the problem is that 'acceptable' means totally different things to different people.  So 'acceptable quality' to someone who's been contact printing large format film will be rather different from 'acceptable quality' for someone who's moving from 35mm film.

That doesn't mean that there isn't an answer.  It just means the answer is complicated, and any final conclusions tend to be couched in terms of sentences that include phrases like "depending on the standard of comparison'.

The Issues

I think there are several things going on with print size. In no particular order, the big issues (in my view) are spatial resolution, noise, and tonal resolution.

 

Spatial Resolution

It's worth noting that with spatial resolution, the MTF of film and a digital sensor differ in dramatic ways. With film, the convolution of the MTF of the lens and the MTF of the film give you a combined curve where as the spatial resolution increases, the MTF drops away from 100% smoothly until you at last are at the zero response at a certain limiting frequency. This means that as the detail you're trying to resolve becomes higher and higher frequency, the apparent contrast of that detail in the capture becomes slowly lower and lower until at last, there's no contrast at all.

This is different from a digital sensor, where the MTF of the sensor is such that it's quite high right up until the spatial frequency approaches the Nyquist limit of the sensor, and then it drops rather swiftly toward zero. There are aliasing and other issues that cloud this generalization but on the whole it's a helpful way to look at the situation.

The result in the print is that you get great fine detail contrast until you're very close to the point where the detail simply can't be resolved, and then the detail just 'disappears' completely. The gradual roll off in contrast is limited to that contributed by the MTF of the lens.

So here's what happens with digital images as you print them. Below a certain size, you're rendering the available detail in the image at a scale at or below the ability of the human eye to resolve - at 'normal viewing distance', this works out to about 7 cycles/mm. (we can argue about whether it's 5 cycles/mm or 10cycles/mm but no one believes it is more like 50 cycles/mm).

Now, by fantastic coincidence, this spatial frequency happens to match closely the 'native' spatial frequency of most high end inkjet prints. The Epsons (like the 9600 I use, or the more current 9800) have a native frequency of 360 pixels/inch, or just a bit over 14 pixels/mm, and Nyquist tells us this corresponds roughly to 7 cycles/mm.

So when we print at native resolution, what we happen to be doing is printing such that all of the detail in the capture is rendered in the print at a resolution the eye can actually detect. (again, more or less.) When we start to scale up the image, we now have a print where the finest detail is rendered at a scale larger than the limit of what we can see, and so we start to sense a lack of 'depth' to the detail. Edges still appear crisp, but areas where we expect to see detail are just smooth tone. [as a side note, it's worth noting that this is only true in the case where the optical resolution of the lens used to make the exposure is high enough that the resolution of the sensor is the limiting factor]

With film, as the actual signal from the film capture passes through this limit, the noise in the image (mostly grain, but also just capture noise in the luminance) starts to be rendered, and provides the 'detail' that our eye craves.

So as you increase print size to the point where you have to up-rez the image, you enter the realm where the appearance of the print from a digital capture and a print from film start to diverge in terms of how detail is rendered. The print from film will have the contrast of the detail roll off gradually, and the noise from film grain and the inherent luminance noise will start to fill in pseudo-detail. The print from the digital capture will have the contrast remain high until the detail suddenly vanishes, and there will be no noise to speak of that comes into play as the print size increases, because there's nothing in the printing path to introduce noise at those spatial frequencies, because they're just clipped off by the nyquist limit of the sensor.

 

Noise

First, a confession. I came to large format because I was drawn by the capacity to render detail and the impressive sharpness. I stayed with large format because I came to love what I call the 'creamy goodness'. The razorlike sharpness and the depth of detail are nice, but what drove me to photograph in large format for so long was, simply, the relative lack of noise.

So, for me, the big win with digital capture is the relative lack of noise. For smooth tonal gradients without high frequency signal (detail) like sky, smooth water, clouds, fog - digital kicks ass.

Furthermore, this noiselessness scales well from digital capture, for the very same reasons that we don't get pseudo-detail like we do from film. When I take a film negative and enlarge it to a certain point, eventually I am rendering the grain of the film at a spatial frequency that's visible. In my mind, this is a 'break-up' of the image, just as some people feel that passing the detail threshold is a 'break-up'. If we have a region of the image that is noise free and has no detail, it doesn't matter how we scale the print - that region will print as noise-free.

Tonal Resolution:

With film, the tonal resolution is limited by noise. That is, the absolute tonal accuracy is limited by the noise inherent in film. We can pick finer grained, less noisy film and developer, but we can't make the noise go away, and the noise floor is the limit of tonal resolution. A lot of the noise floor is the film grain - so that small enlargements have higher tonal resolution than really large ones. Think of film grain as a way to dither a binary signal.

With digital capture, especially the current crop of sensors, the noise floor is really low. I think the 12 bit depth of current sensors surpasses film at all but very small enlargements.

Bottom line

Suppose we were to make two captures of a scene, and print the results both as conventional analog prints and also as digital prints, in a series of increasing size.

If we rated the prints on resolution, at some small print size, they'd be essentially equivalent. That would be the situation right up to some critical size, and then the ratings would drift apart, with the analog print looking relatively better as the noise started to provide 'pseudo-detail' and the digital print looking worse as the details started to look 'blank and smooth'.

If we rated the prints on noise and tonal fidelity, at first, the prints would be equivalent, because the noise in the film capture is rendered at a frequency too high for us to detect, so that it appears as noiseless as the digital capture. As the print size increases, The digital image will remain noise free and will retain all of the tonal resolution, but the film capture will slowly but surely become noisier and have less and less tonal resolution.  As that happens the digital capture will start to outshine the film.

So in practice, what happens is this: at very small scale, film and digital prints are indistinquishable. At a moderate scale, the noise in the film capture starts to be rendered visible, and the digital capture wins. As we increase the scale, the detail in the film capture (and the pseudo-detail) will eventually overtake the noise advantage of the digital capture, and the film will start to win.

 

The Answer

So now we're to the point where I can give a reasonable answer to the question "how big can you print a 5d image?"  Bear in mind, these answers are based on my personal aesthetic.

The answer is something like this: At 10x15, the 5d is beating even medium format film and good lenses, because at that size, MF film would just be beginning to show grain. 4x5 negatives scanned and printed inkjet are only breaking even with the 5d.

At around 15x22, the lack of high frequency detail will start to show in the print from the digital capture, and so a print from a really detail rich source (large format film) will beat the 5d print on resolution but only break even on noise. At this scale, the 5d handily beats the bejeebers out of 35mm film, and edges out medium format film, which beats the 5d on resolution but loses on noise.

At around 20x30, the sparseness of detail is pretty obvious, and FOR MY CURRENT WORK I'd not really want to display prints from the 5d larger than this. Work which is utterly independent of rendering of detail (say, stuff like Stieglitz's cloud equivalents) might well benefit more from the lack of noise and since detail doesn't matter might look great at sizes larger.

That's a very long winded answer to what appears to be a simple question but turns out to be pretty darn complicated. The bad news is that YOUR answer is going to be just as complicated, and because your aesthetic will differ from mine, your answer is almost surely going to be different from mine as well.  The good news is that at least you know how the various aspects of print quality scale as the print size increases.