OMG, says the blogger, the next iPhone’s camera will have 3.2 million pixels instead of today’s measly 2 million! The blog entry gave me the final push for an occasional, meaning at irregular intervals, series of columns on digital photography. The idea is to find insights into what’s really going on in this very dynamic industry, to extract a few useful ideas from the flow of markitecture BS coming from hardware and software vendors on a daily basis. As you’ll see, these columns are intended for the ‘interested’ digital camera user and, on occasion, for the technophobe, but not for the pro – they use cameras to make money, not to have fun like we do.
On to pixels. Moving from 2 megapixels to 4 doubles resolution. No. Picture a square with 2 million dots: that will be approximately 1,400 by 1,400 dots. 4 million will be 2,000 by 2,000. That’s a 41 % increase in resolution, in the number of pixels per linear inch or centimeter, not a doubling. Genuine doubling would be moving from 2 megapixels to 8. Four times the number of pixels doubles the resolution. Going back to the putative iPhone, moving to 3.2 megapixels is a mere 26% increase in resolution, hard to see on a screen or (small) print. When any camera vendor announces a new and improved camera with 10 megapixels vs. last year’s 8, they’re playing us for fools, they know the 12% linear resolution doesn’t make a difference. But, as they say, wait! There is more! The higher resolution could make things worse, it could lead to a decrease in picture quality by introducing more noise. For this we need to go to the sensor.
Picture a matrix, a lattice of tiny silicon photoelectric cells. The technology doesn’t matter here, CMOS, CCD, these acronyms are for the sales pitch. The camera lens forms a picture on the sensor and each cell, each pixel converts the light it receives into electricity, electrons. For each individual sensor, the electrons are measured, counted and the result is reported as a binary number. If the reporting number uses one bit, zero or one, we can only measure the presence or absence of light, that would be a very crude picture. Two bits, four values, 4 bits, 16 values and so on. The higher the number of bits per pixel the higher number of nuances, of exposure values.
The better mainstream cameras claim 14 bits, that’s 16,000 light values. I just used the word claim because there is a fight going on in the sensor, a fight between the electrons generated by the conversion of light falling on each pixel and stray electrons generated by the circuit itself. They’re often referred to a thermal noise; high precision sensors in scientific applications are cooled in liquid nitrogen, inexpensive, or even helium, less affordable but even colder. Going back to the pixel sensor, the smaller it is the smaller number of photons it receives and, as a result, the smaller number of “good” electrons it generates. These “good” electrons get mixed with the “bad” noise electrons and the result is counted. In other words, the smaller the pixel the worse the wanted signal to unwanted noise (S/N) ratio gets. So, for a given sensor size, when you increase the number of pixels, you decrease their size, you worsen the S/N ratio. Such a deal!
This is, of course, a simplification, but not an exaggerated one. Sensor manufacturers know the problem and make every effort to fight the bad electrons, to convert the good ones into bits (A/D, Analog to Digital conversion) before too many bad ones get into the act. But a fact remains: fatter pixels are better. There are two ways to get fat pixels: big sensors and less pixels per sensor.
Another way to look at noise is sensitivity. In the old film days, sensitivity was expressed in ASA or DIN, now in ISO units. Sensitivity, as intuition correctly tells us, expresses how much or how little source light is needed to create a standard image on the target (film and now sensor). On digital film, as on analog one, noise increases as you increase sensitivity, this because you use less and less photons for a given image.
The big fat sensors in some pro or quasi pro cameras reach 6400 ISO or more before noise makes the picture unusable. This is obvious when one manufacturer sells two 12 megapixel cameras; one goes to ISO 3200, the other to ISO 6400. The first uses what is known as an APS/C (the sad story of the APS film format some other day) size sensor of 420 mm2, the other a “full-size” FX sensor, the old 24mm by 36mm format or 864mm2. The FX sensor features the same number of pixels on twice the surface. Fatter pixels yield a better S/N ratio, a higher useable sensibility. (With 3200 or 6400 ISO number, handheld night photography in city streets is now a joy.)
Now, turn to one of the latest Point & Shoot cameras, it offers 15 megapixels on a sensor enigmatically labeled 1/1.7 inch. A little googling gets us a diagonal of 15mm, for a surface of 225mm2 or so. This is about a quarter of the surface for the FX sensor, with more pixels. On this P&S camera, pixels are about one quarter the size of the ones on the big beast; coincidentally, the manufacturer is careful to quote an ISO number one quarter, 1600, of the 6400 trumpeted on the FX format. Things get worse, in the same product line, when 10 megapixels get offered on 122mm2.
Going back to the iPhone, sensor size, not the number of pixels, is what we’ll have to watch for. As for which camera to buy, sorry, no advice here. Convenience, style, size, previously owned equipment, brand pride all complicate the discussion. There is an abundance of very good cameras today and they’re getting better in ways I’ll discuss later, we just have to make sure we know what we’re getting or not getting when we limit the discussion to the number of pixels. —JLG
Next: the color-blind sensor.