Timeline for How can I make a very sharp photograph with a point & shoot camera?
Current License: CC BY-SA 3.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 13, 2012 at 14:28 | comment | added | R.. GitHub STOP HELPING ICE | My point was that it's deceptive to count two optically receptive elements as two "pixels" when, no matter where you point the camera, their outputs are 99.99% correlated due to crap-quality optics essentially band-limiting the picture before it reaches the sensors. Information-theoretic content (e.g. logarithm of the total number of possible images the camera could generate) is the only honest way to count "megapixels". | |
| Jul 13, 2012 at 13:14 | comment | added | Zds | All cameras, also DSLRs, announce megapixels in terms of optically receptive elements. These elements are then filtered by RGBG Bayer array, meaning each element only sees one color. Color information is then (on almost all cameras) interpolated from these samples. | |
| Aug 12, 2011 at 15:31 | comment | added | R.. GitHub STOP HELPING ICE | I would say to be honest they should always measure megapixels in terms of max information-theoretic content the camera is able to capture, factoring in not just the number of sensor sample points but also their correlations due to bad optics, etc. This would put most cheap cameras claiming 10-12 megapixel down around 1-4 megapixel. | |
| Aug 12, 2011 at 14:59 | comment | added | Matt Grum | @R.. cont. use of sensor resolutions not well matched to the resolving power of the lens is a swindle, I agree 100% on that, however the reporting of megapixels is not - how else should they market sensors? Count the number of locations where all three colours are recorded? That'll be 0 megapixels for Bayer sensors. Count the number of RGB triplets required to represent an image of the same level of detail? Well that depends entirely on the image content (a 10 megapixel Foveon image of a grey wall will have the same detail as a 10 megapixel Bayer image). | |
| Aug 12, 2011 at 14:54 | comment | added | Matt Grum | @R.. it's not "severly upscaled", it's slightly upscaled, tests between Bayer and Foveon sensors generally indicate the Foveon (which records an RGB triplet per 'pixel') gives detail equivalent to a Bayer sensor with double the photodiodes (due to correlations the demosiacing algorithm is able to exploit) that's a scaling of about 1.4x. They've always done this so people ought to be used to it by now, it's not a swindle in terms of storage space space as lossy compression is used - if you want the uninterpolated (raw) data that'll cost you more in storage as the compression used isn't as good! | |
| Aug 12, 2011 at 14:23 | comment | added | R.. GitHub STOP HELPING ICE | The "swindle" here is making the "X megapixel" setting on the camera save an image with a total of X million RGB triplets, which is severely "upscaled". It's a swindle in terms of storage space and confusing users and making them think they're getting more than they're actually getting. Of course for cheap cameras there's also the additional swindle that the optical quality is so poor that the individual sample points aren't remotely independent... | |
| Aug 12, 2011 at 13:53 | comment | added | Matt Grum | @R.. All manufacturers, from Hasselblad to Foveon to manufacturers of compacts, state the megapixel count as the number of photodiodes, even if each one is only sensitive to red, green, or blue. If there are 3.6 million photodiodes, then it's a 3.6 megapixel sensor, although some interpolation takes place to produce a colour image. It's not as bigger swindle as it seems given the colour channels are often highly correlated. | |
| Aug 9, 2011 at 1:44 | comment | added | R.. GitHub STOP HELPING ICE | Due to the way the CCD grid works and the oversampling of green, there's a good argument that some such overrepresentation should occur in a rectangular-pixel-grid output, but it's unclear how much, and certainly not as much as manufacturers do. I'm looking for a link to a good blog article I had on it, but can't find it at the moment... | |
| Aug 9, 2011 at 1:41 | comment | added | R.. GitHub STOP HELPING ICE | And easily testable with some trivial frequency-domain analysis. It's well known and documented that manufacturers inflate the megapixels by counting what's effectively (modulo non-rectangular grid arrangement) a 1280x960 CCD as 3.6 megapixel (red, green, and blue! that makes 3 pixels, right? :) then upscaling to a resolution that gives the 3.6 million full pixels. | |
| Aug 9, 2011 at 1:29 | comment | added | rfusca | Thats a pretty bold claim. | |
| Aug 9, 2011 at 1:18 | history | answered | R.. GitHub STOP HELPING ICE | CC BY-SA 3.0 |