Back in the late 80s and early 90s, when I was doing digital photography research, there was a popular topic in the bar at technical conferences: what would it take for digital capture to equal 35mm quality? It was clear that we needed a whole lot more dynamic range; in those days the highly-portable sensors had less than the linear range of chromes, and none of film’s gradual shoulder. But we thought the key issues was resolution. Some made extreme arguments: 200 line pairs per millimeter was the ultimate resolution of some films and many lenses, and that meant 400 pixels per millimeter, or 9600 by 14400 pixels for a 35 mm frame: almost 140 million pixels. Cooler heads, notably those at Kodak, agued that the magic number was 6 million pixels. I remember being at one of Kodak’s Rochester research labs when an engineer slapped down two 20×30 C prints and told me that they were both from a 35mm Ektar 25 negative. One of the prints was an optical enlargement and one was reproduced digitally (on the same photographic paper) using a 6 megapixel file. He challenged me to tell him which was which. It wasn’t easy.
By the time digital cameras became consumer items, the definition of a pixel had changed. We engineers had been talking about pixels that each contained red, green, and blue samples. The pattern used on most digital cameras has equal numbers of blue and red sensors and twice as many green sensors. Information from each sensor is used in combination with information from its neighbors to construct an RGB image. The pixel counts looked much better if you counted each sensor as a whole pixel, so that’s what started appearing in advertisements. You get an argument among engineers when you ask them what the real count should be. I figure that the spectral response of the green sensors aren’t that far from the luminance response of the eye, and that the red and blue sensors serve to add information about color, so it’s fair to throw out the red and blue pixels. So the 140 million pixel cameras that we talked about fifteen years ago would be 280 million pixel devices measured by today’s standards, and the camera the Kodak engineers said would equal a 35mm film camera’s performance would be called a 12 megapixel camera today.
When 11 and 14 megapixel digital cameras hit the market a couple of years ago, photographers discovered that, setting aside the still-unresolved dynamic range issues and a few minor artifacts, the images that they produced were not just as good as 35mm film cameras, they were actually far better. Medium format film cameras flooded eBay as commercial photographers moved from roll film to 35mm-sized digital SLRs, especially the Canon 1Ds 11 megapixel camera.
How could we have been so wrong fifteen years ago? First, we didn’t understand what constitutes quality in a photographic image. We thought that the relative quality of large negatives was the result of greater image detail and didn’t understand that the reduction in apparent grain and the improvement in smoothness was even more important. We could measure resolution, and its more-sophisticated cousin, modulation transfer function, and we fixated on it, ignoring more fundamental and elusive measures of quality. At equivalent resolutions, what digital cameras have over film is image structure: freedom from noise, creaminess of texture, and smooth, evocative edges. That’s because film grain is inherently noisy, much more so than the noise produced by digital capture devices, at least at low sensitivities. I’m amazed when I look at a digital file from a 14 megapixel DSLR that has the smoothness and finesse of a scan of a 4×5 neg (although it lacks the resolution).
The other thing we forgot was how sharpening can compensate for resolution deficiencies. Ham-handed use has given digital sharpening a bad rep, but up to a point, it can mathematically compensate for lens softness. It doesn’t work so well for film: when you sharpen the details in a film image, you’re sharpening the grain as well, and there goes the image smoothness. Digitally-captured files don’t have much noise, so judicious sharpening can yield images with surprising crispness, even if they lack ultimate resolution.