I don’t understand all the fuzz about digital sensor size and resolution versus lens resolution. Higher resolution is NOT (completely) madness! To me the “megapixel madness” meme is “sampling misunderstanding”!
I don’t understand this worry about sensors out resolving lenses wrongly calculated by considering that one (lens) cycle should not be covered by more than 2 sensor pixels...
For example if you have 6.4um pixel size sensor (like my Canon 5DII 21Mpx 24×36mm sensor) you have like (1000/6.4) 156 pixels / mm but this doesn’t mean you need a lens that can resolve at least 156 lp / mm to make good use of the sensor! First lp means line pairs! (a black plus a white line! otherwise how can you see that there are lines!) so that would be ~ 68 lp/mm for the limit where the lens “could” be out-resolved by the sensor (n.b. >= 68lp/mm is ~ a common lens resolution). Moreover lp/mm: well, yes; but at what contrast? the world is not black or white! (Lp/mm is an old crude, non objective measure, now we use MTF measures). To represent a transition from black to white shouldn’t we use more than 2 pixel?! The smallest “detail” a lens can resolve doesn’t exists on its own but only in contrast with another detail (or background) + considering the phase/alignment (to the digital sensor) “it” should therefore be sampled with more than one pixels.
I know, when stopped down, lenses have their resolving power going down due to light diffraction: some think that they are then out-resolved by the sensor (see Why Moore’s Law Does Not Apply to Digital Photography) and anyway many lenses are not that good, but to me a “perfect” – or a good enough – high-end sensor must be able to sample – with more than 1 pixel! – the finest details from what the best lens could get at its best aperture under the best conditions! Not very important for vacation “souvenir” small prints, but for large ones yes!
To correctly record all lens analogical “points” – ~ corresponding to the smallest confusion circle the lens can produce – one must sample them at greater than their resolution (See Nyquist frequency “aliasing can be avoided if the Nyquist frequency is greater than the bandwidth, or maximum component frequency, of the signal being sampled”).
How well a spatial frequency pattern is resolved by a pixel grid depends on the precise alignment (phase) of the pattern with their grid. On average, you can reliably resolve only about 70% of Nyquist; the reduction is known as the Kell factor. In other words, you need at least something like 2.8 pixels/cycle for reliable resolution and reconstruction of a spatial frequency. ... A small amount of oversampling is good, but more is overkill
Camera Lenses: From Box Camera to Digital (via google books)
Let’s say we want 3 pixels to sample a cycle (line pair “section”), and we use a good 80 lp/mm (?at 50% contrast?) 35mm lens, so we need a 240 pixel/mm sensor; for 24mm = 5760 pixel vertically = 5760×8640 pixels = we need at least 50 megapixels before the (35mm full frame) sensor really out-resolves the lens (technicaly that sensor would be out resolving the lens, but this is required for a perfect sampling).
Moreover one must also remember that almost all sensors are bayer matrices (via a color filter array) where half of the pixels are sensitive to green and 25% to red, 25% to blue; their spatial resolution is higher than their color resolution could we say. The real colors are interpolated by demosaicing. I’m sure (high sensitivity, low noise) 50 megapixels for a 35mm full frame sensors wouldn’t be a crazy thing!
(even considering a Kell factor of 0.9 => 2.2 pixels/cycle … you’ll need 27 megapixels to correctly sample the image; with a very high end 100 lp/mm lens you’ll need ~ 42Mpx)
To summary: (35mm) sensors are not yet lens limited.
And if we go to medium or large format: this is interesting. Due to the requirement of a larger image circle, and therefore construction difficulties (design compromises) medium format lenses have less resolving power per mm … so considering a 60 lp/mm lens = (still with 3px for one lp) 180 pixel/mm x 56mm = 10080 pixels horizontally = 10080 x 7560 pixels for a ~ 56×42mm “645” sensor (e.g. the Phase One P 65+) = 76 megapixels (pixel size: 5.6um)...
With the “small” Hasselblad 40×54mm (e.g. H4D-60) sensor you’ll need 70 megapixels (38 if you consider a Kell factor of 0.9… ).
I think the mega-pixels race (and also the higher sensitivity / lower noise + wide dynamic range + high color depth race) is not over! There will be a limit where additional sensor resolution won’t be useful, but we’re not that quite there yet! (but not far) ... For 35mm (full frame) sensor format: 1 more additional (pixel number) doubling would be nice, then halving of the cost, increasing quality/sensitivity would represent a more interesting Moore’s Law. For “small” medium format sensors we’re not far from the limit (with 60megapixels backs)!...
n.b. for the nostalgics: I’ve heard that film is not as crisp as digital, in part due to scattering of light in the emulsion, inter-reflections (= less resolving power, less micro contrast) and higher noise (~ grain)! The resolving power of Fuji Reala (100iso) – my favorite film with fuji nps 160 – is rated 63 “lines” (yes, line widths; black and white, not cycles/pairs or it’s a typo?) per millimeter at a 1.6 to 1 target (weak) contrast ratio = 32 lp/mm, and 125 l/mm at 1000:1 (not a current micro-contrast, more for astronomic photography!?) = 62 lp/mm representing, for 35mm, (125×24 x 125×36) 13Mpx (at 1.6:1 of subject contrast would be 63×24x63×36 = 3.5Mpx). (n.b. those “old” resolution numbers/units look strange … on the fuji provided MTF graph MTF 50% looks like 60 lp/mm : 13Mpx)
To correctly sample that film one would need 1.11^2 times more pixels, but in itself the film represents 4 to 13 millions of analogic “points” and should be for sure spatially maxed out by a ~ 13Mpx digital sensor (nevertheless as color film “dots” can capture any color – no bayer matrix!- film is probably not so bad in “chrominance” resolution, although with its complex layer system, reflexion etc I’m not so sure). I would say that it looks like actual 35mm full frame sensor (e.g. Canon 5DII 21 megapixels) are better than film but not uselessly out-resolving a 80 lp/mm lens …
![three-gorges-dam-project-yangtze-river image [1562px × 1250px]](http://www.mongrelmedia.com/data/ftp/Manufactured%20Landscapes/ML-06.jpg){kind=link}