This is from a recent issue of the Economist. It's as clear an explanation of digital camera sensor tradeoffs as I've read, and it contains some new information as well.
ON RARE and very special occasions, your correspondent digs out his ancient 120-format camera and loads a roll of colour-reversal film from a precious supply of Fujichrome Astia Professional he keeps stored in the fridge. He shoots off a dozen scenes painstakingly composed on the camera’s large ground-glass screen, carefully extracts the film-carrier and removes the exposed spool, wraps it in kitchen foil, and takes it to a lab across town that still knows how to process roll-film properly. Despite the palaver, the silver halide communion is hugely satisfying. Even after all the years of use, viewing the 120's big square transparencies under a loupe on a lightbox can still take the breath away.
In so far as it is possible to compare two entirely different ways of capturing photons, the Hasselblad's 6cm square frame is equivalent to a 70 megapixel digital sensor. The attraction of such a format is that it provides a large enough transparency for art editors to select the crop they like best, while still offering more than enough resolution (when the cropped area is enlarged) to cover losses incurred during plate-making and printing.
Creating a full-page bleed for a colour magazine using even 35mm film is far more challenging. Trying to do so with a compact digital camera or smartphone is out of the question. But, then, the vast majority of digital cameras have light-gathering sensors the size of tiny toe-nails—and are used largely for uploading images to Flickr or Facebook, or for making 3.5 inch by 5 inch (8.9cm by 12.7cm) prints for family albums.
That is not to say film beats digital any day. Not having to pay for and reload a fresh film every couple of dozen shots encourages digital photographers to experiment more. And there is the convenience of being able to see the result immediately, which allows users to delete inferior images and, if necessary, shoot additional ones. Meanwhile, the past decade has seen the light-processing power of silicon sensors become truly awesome.
Depending on the lens and the film speed, a frame of 35mm film has the digital equivalent of between 15 and 20 megapixels. The “full-frame” sensors (with the same 36mm by 24mm format of 35mm film) in digital single-lens reflex (D-SLR) cameras used by professionals can more than match that today. The Canon EOS 5D Mark II, for instance, uses a 21 megapixel sensor.
With a decent lens, even the cheaper D-SLRs produced for the “prosumer” market can come close. Thanks mainly to their smaller sensors (typically 22mm by 15mm), these popular devices tend to be lighter, more compact and less than a third the price of full-frame models. Their sensors are based on the old APS film format, which promised to revolutionise photography but failed miserably. The cameras were barely any smaller than 35mm ones, yet had only 40% the frame size.
So, how come digital cameras that use so-called APS-C sensors, with less than half the sensor area of full-frame cameras, perform as well as they do? Even the so-called “micro four-thirds” D-SLRs, with sensors less than a third the size of a full-frame’s chip, seem more than passable. The answer is that while professional photographers using full-frame D-SLRs may blow up their images to poster size, the majority of camera users rarely make prints larger than 8 inches by 10 inches. Under normal lighting conditions, practically any 12 megapixel D-SLR will suffice.
In fact, 12 megapixels has become a kind of sweet spot in the digital-camera business. With smartphones incorporating eight-megapixel cameras these days, the number of basic digital cameras sold fell 17% during the first 11 months of 2011, according to NPD Group, a market research company based in Port Washington, New York. Meanwhile, unit sales of pricier point-and-shoot models with bigger sensors and zoom lenses grew by 16%, and prosumer D-SLRs were up 12%.
In theory, the more pixels a sensor chip has crammed onto it, the greater is the amount of detail that can be captured in a scene. That was certainly the case a decade or so ago, when Japanese camera-makers were engaged in a megapixel race. But there comes a point at which the pixels are too small and are packed too close together for further advantage to accrue.
At that point, the “shot noise”—caused when individual pixels are so tiny that too few photons strike them—replaces the steady shower of light falling on the sensor with a sporadic, hail-like drumming effect. Meanwhile, the close proximity of the pixels induces thermal and electrical noise. The result is an image that begins to fragment and develop artifacts, especially under conditions of low light.
One answer, of course, is to increase the size of the sensor chip, while making the pixels larger and spreading them out more. The point of using full-frame (and even larger) sensors in professional cameras is not just to increase the pixel count, but also to reduce their density—and thus the background noise. It is the low noise level that allows professionals to produce such detailed images in poor lighting conditions, and to enlarge them to such an extent.
The first rule in choosing a digital camera, then, is to place sensor size over pixel count. A second is to put sensor density and layout ahead of both of the above. Your correspondent has become intrigued of late by Fujifilm’s EXR sensor technology, which uses octagonal pixels instead of the usual square ones. Practically all digital cameras employ a square-grid arrangement of photosensors called a Bayer mosaic, which comprises rows of red and green filters alternating with rows of green and blue filters. This colour-filter array, patented by Bryce Bayer at Eastman Kodak in 1976, has proved extremely successful in capturing both the colour and the detail in a scene.
The use of twice as many green elements as red and blue ones mimics the physiology of the human eye. Because the eye can perceive finer detail in the green part of the spectrum, the abundance of green pixels produces an image that appears to be less noisy and to have greater resolution than would be the case if all three colours were treated equally.
Fujifilm’s EXR technology (originally called Super CCD when first introduced in 1999) produces a higher resolution than is possible with a conventional sensor of similar pixel count. It does this by allowing alternate rows of pixel sites to be read separately, rather like the scanning arrangement on an old-fashioned television set, making a 12 megapixel chip appear as though it were two interleaved six megapixel sensors.
Because the octagonal pixels are aligned diagonally along 45º axes, instead of being arranged in horizontal and vertical rows and columns, pairs of pixels of the same colour can be next to one another, which is impossible in a Bayer rectangular array. In low light, the sensor combines data from pairs of similarly coloured pixels next to one another, averaging out the noise between the two, to reduce the overall noise within the image.
To improve the camera’s dynamic range, the sensor reads alternate lines in the matrix and switches them off part way through the exposure process. This reduces their chance of causing overexposure while still retaining information about highlights in the picture. The underexposed highlight detail is then combined with information from the fully exposed other set of alternate lines, to produce a final image that reveals detail which might otherwise be lost in the shadows.
Fujifilm's recently launched $600 point-and-shoot camera known as the X10 encapsulates all these features and more. And while it will never replace your correspondent’s clunky old Hasselblad, it could well give his even older Leica IIIF a serious run for its money. (Cutting modern 35mm film to form a tongue that fits in the take-up spool of the old rangefinder camera is becoming too much of a chore.) While not quite as retro-looking as its X100 bigger brother, the X10 at least has a bright zoom lens, a faster auto-focus, and an old-fashioned (Leica-like) optical viewfinder. Your correspondent guiltily admits it was this last feature which finally swayed him to make the purchase. Happy New Year!
