Etadam wrote:
So Two 10% inaccurate pixels are significantly more accurate than one 10% inaccurate pixel. Assuming the 2 pixels are twice the size of the \"one\" pixel this is
- either wrong
Perhaps I didn\'t explain myself very well. Imagine that the two pixels together take up the same area as the one pixel, and the micro-lenses are practically gapless. The two pixels should do no worse, in forming an image with low noise, and specifically, accurate colour, than the one pixel. In itself, each of the smaller pixels may be less accurate than the one larger one, but once the two smaller pixels are combined together as an average, for instance by an eye at a normal viewing distance, their combined accuracy should be more or less the same as that of the one larger pixel. The individual inaccuracy of each of the two pixels cancels out somewhat, since the randomness of each of their two noise contributions is independent of the other pixel. That is what I mean by \"smaller pixels can afford to be more noisy in order to deliver the same pictorial result.
That is why it is a fallacy to equate per pixel noise, or colour accuracy (call it what you will), with the effect of noise on the overall image, as you did. Today there is little practical loss of colour accuracy when doubling the number of pixels from say 12 million to 24 million even though each of the individual pixels of the higher density file is less accurate. That is why your statement \"Given a technology, the smaller the pixel, the more likely noise will bias the color accuracy - this is physics\" is quite misleading. The higher noise in each pixel DOES NOT NECESSARILY \'bias\' the colour of the image more, under the laws of physics. Today\'s DSLR sensors are quite close to the idealised sensor, in which there is no noise/colour accuracy penalty for cramming in more pixels, except at the per pixel level, which is relevant only in inverse proportion to the ratio of the number of pixels.
So it is not necessarily the case that \"the smaller the pixel, the more likely noise will bias the colour accuracy\". That dogma is not physics, but a myth whose popularity derives from the widespread failure to view crops at equal magnification. Physics done wrong through a failure to multiply.
Sep 15, 2008 at 09:02 AM
brainiac Offline [X]
Re: Canon 5DII rumors thread
thw2 wrote: Etadam wrote:
Given a technology, the smaller the pixel, the more likely noise will bias the color accuracy - this is physics.
That is common sense.
But a BIG assumption has been made. You are assuming there is NO improvement in technology.
...forget the assumptions: one BIG error has been made. Two 10% inaccurate pixels are significantly more accurate than one 10% inaccurate pixel. So smaller pixels can afford to be less accurate by some margin, in order to produce the same result. \"This is physics\". These days this means that added resolution has very very little downside. The \"keep the pixels few\" crowd are not just ignoring the technological advances, they are also not doing their arithmetic correctly.
If you\'re going to say things like \"this is physics\", it would be a good idea to actually get your statistical analysis right, otherwise it\'s not physics, it\'s fantasy. I\'m afraid that Etadam is still offering a per pixel analysis which fails to take into account the macroscopic effect. Colour accuracy is not a distinct problem.
