Article: How Many Megapixels Do You Really Need in

[BernardLanguillier]

Hi Gary,

Interesting contribution, thank you.

On the other hand, you might want to add some nuance by touching on the 3 following topics:

- Mega pixel is a metrics that was chosen by marketing people in order to help people estimate the quality of a camera. My view is that camera manufacturers do probably benefit more in the long run than customers from the imposition of this sole metrics as the ruling decision criteria.

There are other criterias that as just as important like Signal/Noise ratio (often called noise), color fidelity, sharpness by pixel, Dynamic Range, AF accuracy and ability to focus in low light.

A simple example: a poor AF capability will result in unsharp picture even if the best 8 MP sensor is in the camera,

- Overall, the more megal pixel, the more noise you get, which means that, all things being equal, a 5 Mp camera might deliver better image quality than a 8 Mp camera for some applications where the ISO used would be 400 for instance.

This is a bit nuanced by the fact that newer sensors tend to have better noise characteristics as well as more definition, which tends to equal things out. 8 MP sensor are an execption though as it seems that they are really noisier than earlier 5 MP sensors,

- until now, most sensors were made by Sony and Fuji, but more competition is starting as Panasonic and Sharp are getting more pro-active in the field of sensors. This will make it even less relevant to use MP alone as the metrics, just as it Mhz became less relevant to rate computers as soon as AMD released the Athlon CPUs.

Just my 2 cents,

Best regards,
Bernard

[Lin Evans]

Hi Gary,

It's (megapixels) important, but too much emphasis on "megapixels" can be confusing and too much generalization can lead to faulty conclusions. Let me give a few examples to mull over.

Sports photography in low light. One of the finest sports cameras ever made was only two megapixels and produced publishable images of what might have been impossible for most film cameras. The Kodak DCS-720X gives incredibly sharp and clean images and can be used to get usable frames at even ISO 6400.

The mainstay of sports photography is still done by four megapixel Canon 1D and Nikon D1H cameras. Hundreds of thousands of published and publishable images have been made with these four megapixel wonders. Famed fashion photographer Melvin Sokolsky who shoots for major fashion magazines world wide has used a Canon 1D (4 megpixels) since it became available. In many cases he prefers it to his eleven megapixel Canon EOS-1DS.

I shoot wildlife with a variety of digital cameras (I have over 30 of them) and find some of my very best images come from my Sigma SD10 which many consider to be a 3.5 megapixel camera. I routinely print at poster sizes with results equal or better than I get from my Canon EOS-1DS (eleven megapixels). Of course the Sigma uses a Foveon X3 processor which has 10.3 million photo sensors, but still only produces a file of 3.4 million pixels.

The bottom line is that we need to look at more than "just" pixel count when deciding which camera may be best for our application. Megapixels are important, but optical resolution (not the same as pixel count), noise, sharpness and other factors are equally important considerations.

best regards,


Lin

[Jack Flesher]

Assuming pixel count is most important is simplistic and misleading.

Agreed.  Pixel pitch and overall quality are significant factors.

Given the nature of the "here's an article on my website posts" Gary has made here recently, I assume he's trying to launch his own website on photography -- so I'll add some feedback based on that assumption...

I think the articles are generally well written, but very simple and more like "quick tips" than a real article -- as seems to get regularly pointed out here.  Perhaps you should consider longer, more detailed articles based on these, with these offered as the newbie or beginner mode intros.  Otherwise, you might want to consider posting them in forums where non-photographers hang out (like other hobby forums) where the readers may need some photo pointers and might appreciate the simpler tips.

Just a thought.  Good luck with your site.

Jack

[BJL]

I can recall reading somewhere ... that the human eye could not "see" more than 6-megapixels worth of photo information. ... I notice that theory has been swept aside.

Actually, that statement has not been swept aside at all, once it is properly understood. For prints viewed so that they fit into the so-called normal field of view, viewing distance no less than the diagonal length of the print, a good 5MP or 6MP is enough to match the resolving power of most human eyes. Pixel counts going well beyond 6MP serve three purposes:

a) to allow for large prints that can handle close scrutiny of part at a time, like an A3 or 16"x20" viewed from a typical close viewing distance of 25cm [10"].

 to allow for substantial cropping (useful for example to people who prefer to spend extra money on equipment rather than thinking about composition or trying several different framings of the same subject.)

c) to dazzle ignorant customers (but apparently even most P&S digicam buyers are not fooled anymore; they are realizing that 5MP is as much as they need, and so marketing is moving to other attractions, like size.)

[Mark D Segal]

Lin,

I appreciate your erudite discourse on this issue - I expected feedback on that post! Now let me ask you: what is your definition of "optical resolution"? That could determine whether we are thinking similarly or not, in the final analysis. Remember, disaggregating an issue doesn't make the issue simple - it just makes it more tractable to deal with; and yes to some extent I appreciate how the behaviour of certain variables can have some mutually off-setting or re-inforcing impact on image quality, but I also think that at some point there are determinate relationships between PPI and print size you move beyond at your own risk no matter what the other variables.

Of couse, I was not including any consideration of film in my post, because that would be comparing applies and oranges. I scan plenty of film at OPTICAL resolutions exceeding what I can get from my Canon 1Ds and the Canon 1Ds still produces higher quality images with much less post-capture work - the grain issue being they key factor.

Mark

[Lin Evans]

Yes, I was referencing optical resolution as measured by resolution chart photos in either lines per image height or line pairs per mm, but of course that's excluding some relevant factors simply because resolution charts are black and white and there is the issue of color resolution which is not measured but still relevant.

For much of what we consider "resolution" there is a somewhat linear relationship between sampling site count and optical resolution, but we must segregate sampling site count from "pixels" in the relevant sense because there is not "always" a direct one to one relationship. As an example, the Foveon X3 uses three sampling sites to create a single pixel where bayer processed sensors ultimately create a pixel per sampling site even though the data is not directly gathered but rather a degree of it is estimated from adjacent values. But even though the Foveon X3 sensor only ends up with a display pixel count of 3.4 million pixels, the black and white optical measurements are roughly equivalent to those measured by a six megapixel bayer equivalent.

So to use "only" a pixel count can be deceiving. So it is with the differences in optical resolution between something such as an eight megapixel (sensor site count) consumer grade sensor and an eight megapixel pro grade sensor such as found in the Canon 1D Mark II.  There can be as much as a 200 line difference in their optical resolution count as measured by resolution chart photos, yet their pixel count may be virtually identical. The difference being the quality of electronic processing and the number of electrons captured by the photosite wells which render better signal to noise ratios.

The important issue then is how accurately these photosite samplings represent reality and how many sampling sites are available to describe the boundaries of fine detail. The better the detail is rendered, the more accurate the interpolation and the greater the enlargement potential.

Some day perhaps we will have sufficient standardization that we can simply relate a pixel count to print size potential for a given sampling density at a particular angle of view, but it may be a while before that happens.

Best regards,

Lin

[BernardLanguillier]

This discussion can be further extended by the introducing the related notion of accutance (or micro contrast) which also impacts a lot the feeling of sharpness of an image.

Regards,
Bernard

[Ray]

Choosing a camera based solely on pixel count is like choosing a wife based solely on bust size.

By the way, choosing a wife on boob size could be very nice for the kids. Do you want happy children or not?  

[gary_hendricks]

Here is an article I published on my website about the number of megapixels needed in a digital camera. I hope it'll be useful to some of you in this forum.

How Many Megapixels Do You Really Need in a Digital Camera?
by Gary Hendricks

If you're out buying a digital camera, then one of the key deciding factors is the number of megapixels supported. The number of megapixels determines the how good your photos turn out. If you have too few megapixels, then your pictures will turn out crappy. Investing in a camera with too many megapixels, on the other hand, is an unnecessary waste of money.

Personally, I feel that if you have unlimited funds, then, by all means go for that high end 8 megapixel camera. Otherwise, you certainly don't want to waste money on extra megapixels you don't need.

A golden rule to bear mind: a camera with more megapixels isn't always better. If your camera supports more megapixels, then each photo you take will be larger. This means that you'll use up more space on your memory cards and computer's hard drive.

If you have trouble deciding how many megapixels you need (I know I did when I bought my first digital camera), then the guide below will help. Essentially, you need to ascertain what size prints you want to get and what your budget is, before deciding on how many megapixels you want. So here we go:

1 megapixel or less: Cameras in this range (e.g. web cameras  or cell phone cameras) have very low image resolution. Don't expect to be able to print high-quality photos using these cameras. You can, however, email the photos or post them on your web site. The good thing about such cameras, of course, is their low price.
      
1 to 2 megapixels: My first digital camera was a Canon PowerShot S110 which only had a 2 megapixel sensor. Cameras in this range are pretty decent though - you can expect to print out great 4x6 prints at this resolution. Of course, if you want larger, blown-up protraits of your birthday party or holiday in Italy, then I would certainly recommend getting more megapixels. Cameras in this range should sell for around $100 currently.
      
3 to 4 megapixels: Most new point-and-shoot cameras these days tend to have at least 3 to 4 megapixel image resolution. Bring these images to the lab and they'll be able to develop great looking 4x6, 5x7 and even 6x9 printouts. Expect to pay slightly more though - we're looking at around $250 for a good model.
      
5 megapixels and up: The more advanced cameras tend to have image resolutions of 5 to 8 megapixels. Newer point-and-shoot cameras have 5 megapixels, while the newer digital SLRs come with 8 megapixels. The quality of images shot by these cameras is simply stunning. Of course, their price tags are equally stunning . In this megapixel category, expect to pay around $300 for a 5 megapixel camera and up to $1800 for an 8 megapixel SLR.
      
Conclusion
Well, now you know roughly the number of megapixels you should be shooting for depending on your intended usage and budget for the camera. My general advice is, if you're just an amateur photographer, then don't buy cameras above 5 megapixels. When you are really serious about digital photography and want to go professional, then consider buying a super high megapixel camera.

[Jonathan Wienke]

I agree with Bernard, megapixel count is only part of what one should consider when buying a digital camera. Autofocus speed and accuracy, viewfinder quality, sensor S/N ratio, color accuracy, buffer size, lens options, dynamic range, shutter lag, maximum desired print size, and per-pixel image quality are all factors one should consider. Assuming pixel count is most important is simplistic and misleading.

[HiltonP]

I can recall reading somewhere (I wish I could remember the source) about two years ago that the human eye could not "see" more than 6-megapixels worth of photo information. Of course this was at the time that 6-megapixel dSLR's reigned supreme. I notice that theory has been swept aside.

[Bobtrips]

I can recall reading somewhere (I wish I could remember the source) about two years ago that the human eye could not "see" more than 6-megapixels worth of photo information. Of course this was at the time that 6-megapixel dSLR's reigned supreme. I notice that theory has been swept aside.

That's going to depend on the size of the print.

When you get to print size it's better to think of pixels per inch (ppi) rather than the number of pixels.

My two meg camera has a 1200 x 1600 'array' of pixels.  I can make a 4" x 6" print at about 260 ppi and I doubt that many (over the age of ten) could tell the difference between a print from that camera and one made from a 6 or 16 meg camera (in terms of resolution).

But at 8" x 10" the difference between a 2 meg and a 6 meg capture are quite easy to see.  Compare a 20" x 30" print made from a 6 meg digital and a 16 meg digital and you should have no problems telling them apart.

[Mark D Segal]

I like to "disaggregate" the issue this way: ask yourself what is the largest print you would make at the lowest PPI you consider acceptable and the smallest print at the highest PPI for maximum visible quality, making allowance for the fact that a certain percentage of pixels one captures could be sacrificed to cropping for compositional purposes. Given max/min size, min/max PPI and an allowance for cropping, the maximum megapixels one needs is mathematically determined. Then one can select a camera within that MP range depending on all the other important quality factors mentioned in other posts.

[Lin Evans]

like to "disaggregate" the issue this way: ask yourself what is the largest print you would make at the lowest PPI you consider acceptable and the smallest print at the highest PPI for maximum visible quality, making allowance for the fact that a certain percentage of pixels one captures could be sacrificed to cropping for compositional purposes. Given max/min size, min/max PPI and an allowance for cropping, the maximum megapixels one needs is mathematically determined. Then one can select a camera within that MP range depending on all the other important quality factors mentioned in other posts.

Hi Mark,

Unfortunately, it's not that simple and can't be reduced to absolutes based on pixel count and math.

Digital imagery as relevant to digital cameras absolutely depends, in the final analysis, on optical resolution. Regardless of pixel count it's what the sensor can or can not capture which determines the printed outcome.

With film, we simply know that with a particular grain, film type and platform we can print at such and such a size with practically guaranteed outcome. We might say that enlargement potential is grain limited simply because we can't print large enough because of grain intrusion to reach resolution limits. Digital, on the other hand, has such low noise (digital grain equivalent) that we can quite easily reach resolution limits while still producing very clean output.

Resolution limits depend ultimately on optical resolution. Optical resolution does not necessarily depend on pixel count. My Sigma SD10 only produces a 3.4 megapixel count file, yet has the optical resolution of a six to eight megapixel Canon or Nikon equivalent.

In the past we routinely printed huge head and shoulders portraits made with our D30 (three megapixel) Canon. These prints were beautiful, crisp and had detail and quality far better than we were ever able to achieve with fine grain color film in 35mm platform. With film we simply ran up against the grain barrier at around 16x24 inches. On the other hand, with fine grain film we could depend on printing beautiful wide angle, hyperfocal landscapes at 16x24 while our D30 struggled with an equivalent 8x12. The difference? Simply the amount of optical resolution. Because a head and shoulders portrait contains much less "geography" than a wide angle, hyperfocal landscape, the D30 had more than sufficient optical resolution to allow huge poster sized prints. But to properly define the landscape and print at 16x24 with results equal to out best attempts with 35mm fine grain film we need our Canon 1DS (11 megapixels).

The essence is that all digital prints other than the one single size which equates to horizontal by vertical matrix divided by desired pixel density in PPI requires interpolation. Interpolation algorithms work nearly perfectly to render accurate enlargements "IF and ONLY IF" they are provided with real data which means properly defined detail. If they get garbage in they put garbage out. Improperly defined detail is commonly referred to as "marker pixels." Marker pixels are place holders which are quite sufficient to fool our eyes and brains as long as they lie beneath human visual acuity thresholds. We easily accept brush strokes in oil as pine needles on distant pine trees until we look at the oil painting under intense magnification. At such times the illusion is destroyed and our senses are forced to see brush strokes in oil.

When marker pixels are encountered by interpolation algorithms they are simply enlarged and during the print or on-screen display process are raised to the threshold of our visual acuity. Should we step back a few feet, our brains once again accept these marker pixels as sufficient representations of detail we "expect" to find. Come forward or magnify the image and again we are forced to see reality.

There are always marker pixels, no matter how much resolution we have in digital or film. Where they "begin" depends on optical resolution. The more optical resolution the more geography we can handle within the frame and the more we can enlarge before we run up against these issues.

The bottom line is that optical resolution "and" frame geometry both need to be considered when shooting digital. It's not as simple as reducing it to a pixel count for a given print density..

Best regards,

Lin

[Jonathan Wienke]

Choosing a camera based solely on pixel count is like choosing a wife based solely on bust size.

[Mark D Segal]

Jonathan,

Just beautiful - I would commit this to memory if it weren't already far too late - drat. I have chosen both already - one some decades ago and the other much more recently. I'll leave which is which to your impeccable deductive logic - and just so you'll know, niether bust size nor pixel count were uniquely determinative variables in either case.  

Lin,

OK, I see where you are coming from if one is thinking of optical resolution in terms of the lp/mm one can read from one image versus another, and I do of course accept that two different 8 MP cameras can produce different image quality as function of the variables you mention - which is why I never proposed one should select a camera on the basis of MPs alone. But absent any post-capture rezzing-up, I would still expect more visible evidence of pixellation on say a 20 by 30 inch print from a Canon 20D than I would from a Canon 1Ds or 1Ds MarkII.  That is why I think regardless of compensating factors the relationship between pixel count and image size is neither determinative nor immaterial. Useful discussion - thanks.

Cheers, Mark

[Ray]

All this confusion results from the fact we don't have MTF information for most lenses and, it appears, any MTF equivalent for any sensor.

The quality of a digital image is ultimately dependent on 3 things; the MTF response of the lens at resolutions that are relevant to the 'pixel density determined' resolution limits of the sensor; the 'quality' of the pixel or photodetector; the total number of pixels.

Of course there are many other camera features which can contribute to the resolution of the final image, which is why I use the word 'ultimately', ie. having taken care of those other factors such as tripod and proper focussing etc.

Let me expand a bit. Digital sensors tend to have a sharp cut-off point with regard to resolution. For the 10D it's about 54 lp/mm. Doesn't matter how good your lens is, you're not going to get more than 54 lp/mm (Okay, maybe 56; let's not quibble).

Anyone designing a lens specifically for the 10D, could safely ignore all design effects that might reduce performance beyond, say 50 or 60 lp/mm. The goal would be to produce a maximum MTF response at all frequencies up to 50 lp/mm.

MTF at 50 lp/mm is unavoidably less than at lower frequencies. That means the signal is rather weak and tends to get obscured by the noise of the sensor. We can address this problem by increasing the MTF response of the lens at 50 lp/mm, and/or reducing the noise factors the pixels are exposed to; or to put it another way, increasing the quality of the pixels.