How does 18 megapixels compare to 12 megapixels?
I'm purchasing a digital camera & it says, "Effective pixels: Approx. 18.00 Megapixels".
What is this compared to 12.00 megapixels?
Don't get too hooked up on megapixels. Manufacturers would have you believe they are the be all and end all of quality, when they're really not. They simply determine how big you can print the image. To give you an idea, a 6MP image could be printed 30" wide and still be perfectly acceptable from a couple of feet away - a 5x7? Perfect even close up. You could probably print a billboard with 18MP - but how and why would you do that? So, don't let a salesman convince you to part with another thousand bucks on the strength of more megapixels. Get a better lens or two instead.
@ElendilTheTall: I agree, but I also gotta add that more megapixels are also nice because, assuming the picture is taken so to take actual advantage of them, it gives you more room to crop — you can print details rather than the whole frame.
Nice, but not vital. I'd rather have 3 lenses and a 12mp camera than 1 lens and 18mp.
"Effective" is an important word here, and probably means there is some trickery going on in the camera that the manufacturer claims is going to make it "like 18 Megapixels". Take that warning on top of all the others about how Megapixels don't necessarily make for quality.
That means 22.5% more pixels in each direction.
5196 x 3464 instead of 4242 x 2828. (The megapixel value is of course rounded, so the exact resolution varies between cameras.)
If the cameras are otherwise comparable, you get more details with highter resolution. However, 12 megapixel is good enough for most uses. If you for example make 4" x 6" prints, you can't see the difference between 12 megapixel and 18 megapixel. If you enlarge a part of the image, the higher resolution will allow you to enlarge it more.
Indeed, if you make 4x6" prints, you probably won't see the difference above about 2 megapixels (1800x1200) as that is 300ppi.
The resolution, by itself, doesn't mean much. For a meaningful comparison, you need to look at one camera versus another, not one resolution figure compared to another.
Assuming you're talking about an APS-C sensor, 12 MP is high enough that the sensor resolution is only rarely the limiting factor on the resolution of the picture you get. In a typical case, the lens (and your camera technique) will reduce the resolution below what the sensor can (theoretically) produce. That being the case, a higher resolution sensor will only rarely help much.
At the same time, I should also point out that if you do have at least one really good lens and (for example) use a tripod at least part of the time, there's a fairly reasonable chance that when you use that combination an 18 MP sensor really will produce pictures with slightly higher "real" resolution than a 12 MP sensor.
It's also true that higher resolution in a fixed size of sensor is likely to imply smaller sensor wells, which will typically reduce light collecting ability. In theory that could lead to higher noise. Of course, there's a lot more to noise level than just sensor well size though. The difference in sensor well size is not enough to say with any certainty that a camera with an 18 MP sensor will have higher noise than a camera with a 12 MP sensor (in fact, if the 18 MP sensor is newer, noise might well be lower). If all else was equal, the 12 MP sensor should have lower noise, but all else is almost never equal.
To get a meaningful comparison, you really need to compare one camera to another, not one sensor resolution to another. Sensor resolution is a factor to take into account, but it's only one of many, and it's not (IMO) a particularly important one either.
Edit: Okay, after supper tonight I had a few minutes, so I wrote up a simulator. I'll add a few pictures here:
Lens resolution exceeding sensor resolution:
Sensor resolution slightly higher than lens resolution:
Sensor resolution much higher than lens resolution:
As someone who owns the 12.2mp Canon 450D, and who has used a friends 18mp Canon 550D on several occasions, I can say with confidence that the newer 18mp sensor has far superior noise performance. Noise on the 450D is terrible beyond ISO400. Noise on the 550D is acceptable up to ISO1600. I would say that sensor well size is a minor factor in noise these days.
@jrista: yes, I was trying to say that could be the case, but after rereading it decided to edit to emphasize it more.
It's worth noting that from a technical standpoint neither the lens nor sensor are ever solely the limiting factor in the resolving power of the system. The spatial frequency response of the system is the product of the spatial frequency response of the lens and sensor, meaning that even if you have a "soft" lens, it will still produce sharper images with a higher resolution sensor and will continue to do so as sensor resolution increases, though by smaller increments.
@Matt, can you elaborate on that? The way I think about it, if the lens produces spacial frequencies up to X Hz, and the sensor resolution sampling is such that its nyquist is above X Hz, I don't see any benefit to oversampling the incoming lens image. Is this the wrong way to think about it? Or even thinking of it from optics, a higher resolution will just mean a smoother airy disk, not a sharper airy disk.
@Eruditass: technically, Matt's right. If you think of an airy disk, what's being projected by the lens is is some specific size of illuminated circle. Being circular, that will never coincide exactly with a particular set of pixels in the sensor. Higher sensor resolution comes closer to an exact circle, and the smaller of pixels at the edges that are partially exposed, so the cutoff at the "edge" is sharper.
@Jerry, when I go http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm and look at f/16 ( simulating a low quality lens) and go from the lowest pixel area and move to smaller pixel areas, it seems to me that the 1D will produce the most perceived sharpness, the rest will begin to bleed into surrounding pixels more and more and decreasing the contrast between the peak and the surrounding black pixels. Now when there are spatial frequencies above the nyquist or anti-aliasing filter frequency, I know it's a multiplication of each component's frequency responses.
@Eruditass: At least IMO, that page doesn't do a very good job of showing what you'd really see. I'll have to think about it for a bit, but perhaps I can throw together a somewhat better simulation. It *is* true that in some cases where the size of dot coincides (almost) exactly with a pixel that a larger pixel will give a cleaner edge -- but in reality, this happens so rarely it hardly matter.
@Jerry, I'd like to see a link or argument that definitively shows that increasing sensor resolution has a benefit in an airy-disc limited situation. I'm skeptical that there is a better simulation than something as simple as that. It would be the same to say that when diffraction limited, higher resolution is still better. So, a 20 megapixel P&S really is sharper than a 12 P&S megapixel at f/8, all else equal.
@Eruditass: I wrote a small simulator tonight, and edited some screen shots of it into the answer.
@Jerry Coffin: simulator = cool. Can you add some pictures showing specifically 12→18 mpix?
@mattdm: I'll have to do some more work on it before I can do that. Right now, the circle size and the sensor grid work in terms of pixels, without trying to relate that back to specific sizes in the real world.
Interesting simulation. A couple criticisms: the airy disk is not uniform intensity, but rather something like a sinc squared function. Second, the difference between the the light and dark pixels will be the greatest in the 1st case, where most of the photons are counted in the single well and not spread out. And we are talking about perceived sharpness, which is best measured by modulation transfer function, or perceived the spatial frequency, directly related to contrast.
@Eruditass: Anybody who's looked at the bokeh produced by various lenses realizes that the distribution of energy does *not* follow a single (or simple) function. IMO, to really improve on a uniform disk, the minimum you'd have to simulate would be diffraction, spherical aberration, coma, and astigmatism. To do very well, you'd probably also want to simulate the color filters of a Bayer grid, and add in chromatic aberration. Unfortunately, we don't usually have access to the parameters (e.g., index of refraction of each element in a lens) necessary to even begin most of that.
As far as light versus dark pixels goes: yes, a bigger pixel will give the highest contrast, but it (obviously enough) also gives a bigger pixel. If you'll pardon a reductio ad absurdum, a single pixel the size of the whole sensor wouldn't produce very sharp pictures -- and even if you increase that to a few hundred, it still doesn't look very sharp.
@Jerry, Yes, it will give a bigger pixel, but of course we are debating over whether it matters when the sensor already out-resolves the lens. At this point, there is absolutely no potential for spatial-frequency gain in MTF, only contrast gain, if at all.
Also, I'd say a point source in the area of maximum focus in a low quality lens with weak correction of the aberrations you mention is much more akin to a sinc^2 or gaussian distribution than a uniform distribution like bokeh. Bokeh happens because the plane of max convergence is before or after the sensor plane, and the rays are approaching linearly as there are no elements between the last element and the sensor. Completely different situation.
Yes, there is a potential for the spatial frequency to increase -- by roughly half a pixel (on average) at each edge. Perfect focus across the frame is a rarity except (maybe) for those people who only take perfectly focused pictures of completely flat objects.
@Jerry, how is there potential for the spatial frequency to increase if the premise of the situation is the lens is limited in resolution and the only thing changing is the sensor megapixel count? a 20MP P&S really is *ALWAYS* sharper than a 12MP P&S and never equivalent? The only way for this to happen is if the sensor's effective fill factor, AA filter, etc. are such that the sensors MTF is greater at those frequencies that the lens displays. This is **NOT** inherent in just increasing the sensor's megapixels.
@Eruditass: if light hits any part of the pixel, the whole pixel gets that much exposure. Even if the focused spot is small enough that it *could* all hit one pixel, it may not have the courtesy to do so -- it could hit up to four instead, and in that case their being smaller reduces the size it's shown as in the picture.
@Eruditass: Much of your claim seems to be based on the notion of a spot of light hitting only a single pixel, and therefore producing a sharp cutoff at the edge of that spot. In reality, that's only rarely the case. I think, however, that this conversation has reached (and probably gone beyond) the end of its useful life.
@Jerry, it then appears that you are misunderstanding my argument: and understandably frustrated. I'm a bit shocked that you feel my points are based on that when the very notion of a gaussian or sinc^2 distribution over the pixels is contrary. The premise the situation that we are discussing is that a spot of light will not fit in one pixel (limited in spatial resolution by incoming light).
I believe you have a misunderstanding as to what an increase in spatial frequency and perceived sharpness means: *adding pixels on the edge at lower intensity (from a higher sampling rate) will maintain the same spatial frequency*, as it is all coming from the same band-limited source. In signal processing, higher sampling rate means nothing with a band-limited source, all else equal. If you do a 2D FFT, it will be the same, with the same sensor MTF. Check out http://www.normankoren.com/Tutorials/MTF.html and DSP.
In fact, by the nature of the problem (lens is of low quality e.g. band-limited spatial frequency), you simply can't have a a uniform disc in the way you describe. If it drops off as fast as the pixel spacing, it isn't more limited in spatial frequencies than the sensor. By definition, it needs to drop off such that the edges do not contain higher frequencies than the pixel spacing. You can't create something out of nothing. I'm convinced you are arguing against a different problem given your assumptions on my argument, but encourage you get on the same page about the situation we defined.
@Eruditass: if you really want to continue, perhaps you could start by reading: http://www.zeiss.com/C12567A8003B8B6F/EmbedTitelIntern/CLN_30_MTF_en/$File/CLN_MTF_Kurven_EN.pdf, especially starting around page 10.
There are two reasons in my mind why you might care about more pixels, one of which might apply to the average photographer, and one that would apply mostly to pros:
Retouching. Pro photographers often use a medium format for the super critical mega dollar shoots. In these situations, a Phase One digiback (39 megapixels) or a Hasselblad (31 mp) or some such would be typical--even when the final print will be magazine size. The reason why so many pixels are useful is for the post-process folks to have tons of extra data. When you zoom in at the micro-detail level to remove a wrinkle in the models face, the photoshop guru's can make a really convincing coverup when there is lots of pixels to blend (or whatetever they do--I have't a clue what tricks they use). Basically having a ton of extra detail pixels allows for more seamless retouching. Most mere mortals never have to worry about having this excess of pixels
High quality prints. There are numerous posts on the details i/r/t getting the highest quality prints out of an inkjet. To simplify, basically for me I print 360 lpi. That means that my 18mp images are natively printing at about 9.3x14.9 inches at that print resolution (note I'm saying NATIVE--no enlarging). High Quality = magnifier close. None of the namby-pamby 'diagonal viewing distance' stuff (I personally think that whole diagonal viewing distance rule automatically takes you out of the 'high quality' realm--even if it's perfectly fine for the application--it's just not the highest quality). Some people report that you can get as good of quality as native by enlarging. I haven't tried so I couldn't say, but there is surely a limit to how much you can enlarge before the image will be degraded so let's just for arguments sake say that for an 18mp image, the max high quality print size you can achieve is like 16x20" (again this is just a rough estimate).
A 39mp image printed natively at 360lpi would give you a 13.7 x 21.9" print. That's big, but not huge. So even with this lavish excess of pixels, you can see that if you are going to turn the quality up to 11 you are not printing billboards by any means. Even if you find that there is no difference by enlarging up to say 50% you can only get to poster size with 39mp--you aren't going to get a high quality poster out of 18mp. On the other hand, if you can stomach lower quality (and don't kid yourself, it's lower quality. It might be acceptable quality, but it's lower than maximum) then you can print as large as you or your client can accept.
A bit of an update is needed here: the current top-of-the-line Phase One back (the IQ180) is full-frame 645 and 80MP; Hassy is now sitting at 60MP (using the same full-frame 645 sensor as Phase One's previous leader, the P65+). When I win the lottery (and I *will*, although the not buying tickets thing makes that a little harder) I'm definitely getting one or the other.
A very rough guesstimate: 50 per cent more...
@decasteljau so is this answer - 18 definitely compares to 12 by being 50% bigger!
It's more like 22.5% more, as Guffa said. It's 50 per cent more *area*, but 22.5% more resolution in each direction. Kinda like how you'd have to go from 6 megapixels to 24 megapixels to double the resolution.
22.5% more pixels in each dimension, but the _resolution_ increase is a more difficult question.
If you are asking this question, then you do not need 18mp :) I have a 450D and my friend recently purchased a 550D. We have done several 100%+ totally unnecessary pixel-peeping comparisons of real world scenes on tripods with physically the same lens, and the resolution increase is noticeable but not incredible by any means. For you to actually see a visible difference in a print, it would have to be upwards of 20x14".
That being said, the noise performance of the 550D vs 450D is far more noticeable than the resolution. Yay for Digic4. Overall handling is nicer, and movies are fun to play with. However if you are just comparing 18mp vs 12mp, I would considering it null, buy whichever camera has other things that you want and ignore mp. 12mp is more than acceptable for probably 95% of all pictures taken in the world, 95% of which, probably will never see print anyway, so 3mp would have been enough :D
A camera with more megapixel's should take more detailed photos than one with less, because the resolution of the camera is higher in the same way a HDTV will look more detailed than a regular TV because the resolution is higher. These higher resolution photos can be printed larger or cropped to a further extent without looking blurred.
The 18megapixel camera should be the better camera (especially with it having a 50% higher resolution) however there are other factors in play such as quality of the lens, the sensors sensitivity to noise that will affect the quality of photos. This is much the same as engine size is a reasonable way to measure a cars top speed, it makes the majority of the difference but factors such as weight, quality of tyres will also affect the speed.
not necessarily. Packing more sensor elements into an array of a fixed size isn't always a good idea (and with APSC/DX size sensors, anything much over 10MP is going to hit the limits of what makes sense, anything higher is just a marketing gimmick).
@jwenting: That is a good point, however the size of the sensor also makes a difference. With a larger sensor you can get a higher resolution before you reach the physical limits.
However, the 550D is an APS-C, not a full frame. 18MP on that size sensor is asking for noise.
@jwenting Whilst you will start to get diminishing returns with megapixels for a fixed size sensor I don't think we've got there yet with APS-C, there is a clear difference in quality between a 10 and 18 megapixels sensor.
I agree with you in principle @jwenting, but there have been advances in sensor tech on the 550D as explained here: http://www.dpreview.com/previews/CanonEOS550D/page3.asp . Either way the nature of the original poster's question suggests they don't have a great understanding of camera technology, hence my simplification.
For the vast majority of users there's no "clear difference" between even a 6MP and a 10MP sensor, let alone an 18MP one. They shoot their pictures, maybe tweak a bit in the software provided with their camera, and get them printed at 10x15cm or smaller. At that print size, even 6MP is more than enough (I until recently had to reduce the DPI of my images from my 12MP D200 as the full size files would literally crash the digital printing machines, says enough? )
Sensor tech aside, it does take more work to really get more resolution out of a higher-density sensor. Your focus has to be more accurate, and you're more susceptible to camera-shake blur.
@jwenting My experience is that "the vast majority" of people don't even print their pictures, they look at them on their computer, with or without adjustments. In which case even 6MP vs 18MP makes *no* difference because even a large monitor is only on the order of 2-2.5 MP. (For example, 1920x1200, which is the highest pixel count I've seen in regular use, works out to 2.3 MP.)
@ElindilTheTall: I've used the 550D on several occasions, and I own the 12.2mp 450D myself. I can say that the noise on the 550D is FAR better than it is on the 450D. I can barely go above ISO400 before noise becomes a real problem...on the 550D, ISO1600 was very acceptable. Noise is no longer simply a factor of well size...electronic advances in photosite design have made significant gains regarding noise performance, and higher resolution no longer automatically means worse noise.
@jwenting @Michael I have observed many people representative of the vast majority of users, and most of them don't print at all. What I have seen them do on a regular basis is load the images on their computer **a zoom right in!**. They zoom in on group photos to look at individuals funny expressions, they zoom in on weird looking stuff in the background and them zoom in on the computer when the subject was too far away for their optical zoom. And with this behaviour there is a big difference between 6 and 18 megapixel sensors noticeable to your average user!
Unless you are willing spend a considerable lump of money for the LENSES you will not notice any difference at all. As Guffa already demonstrated, 18 MPix gives you roughly 22% more pixels in either direction. However, if you want to max out the 3464 pixels available vertically you need, as a rule of thumb, a lens which resolves 1700 l/p (line widths/picture height) at MTF50. A (very) good kit lens usually resolves between 1000 and 1200 l/p, even more expensive lenses seldom reach 1700 on an APS-C sized sensor. So, on anything that is smaller than a full frame sensor, 18 MP effectively does not produce much usable additional detail at all.
If you want to see it with you own eyes, then you should have look at dpreviews's exhaustive reviews which feature resolution chart tests for every camera. Here's the resolution of a 18 MP Canon 550D:
compared to the Canon 450D:
-1 You don't necessarily need a good lens to take advantage of extra megapixels. When people measure MTF values using charts they are measuring the MTF of the system (lens + camera). The MTF of the system is the **product** of the lens MTF and sensor MTF. This means that if you improve one the sytem MTF improves even if the other remains constant. So increasing the number of megapixels (assuming this increases sensor frequency response) will increase the resolving power of the system even with the same lens.
Also the colour resolution of bayer sensors lags the monochrome resolution, so having more pixels reduces Bayer artifacts and gets you closer to the output of of a full colour Foveon type sensor. Finally the links you posted contradict your statement as they show the 550D resolves finer horizontal detail (aliasing starts to kick in around the 26 mark) than the 450D (aliasing can be between 22 and 24).
If you are jumping from a 12mp full frame camera like the Canon 5D to the 550D, the only advantage you'll get with the 18mp is a larger printing size. I have both cameras and can say the 5D still offers better sharpness, detail, and depth. The 550D offers gorgeous colors and great noise handling, a little better than the 5D, but those things can be tweaked in the computer anyway. I still prefer the 5D not only for the image quality but for its robustness.