Megapixels - How Many Is Enough?

The resolution and size of a digital camera's sensor are two of its most fundamentally important characteristics. Assuming that a camera has a high quality lens, the sensor is what ultimately determines the camera's potential for image quality. In conventional photography the corollary is the camera's film format. Digital photography is in the midst of a rapid evolutionary process and there are few (if any) format standards. The closest we have is the Megapixel (Mp) rating. But as you will see, it is a standard that can be somewhat misleading.

It would be easy enough to make a simple chart that shows how many Megapixels are needed for a particular sized enlargement or use. In fact, you can often find such charts and information prominently displayed where digital cameras are sold and even on stickers attached to the cameras themselves. But are these estimates reasonable? This article will help you answer that question for yourself and give you some background that will help you to assess digital cameras as they continue to evolve.

What Is a Megapixel?

The horizontal sensor count multiplied by the vertical sensor count and then divided by one million is the camera's Megapixel rating. At a basic level, the greater the number of individual sensors the greater the potential detail in an image that it creates. Most digital cameras have sensors ranging from 2 to 5 Mp. These are typically marked in 1 Mp increments. All things being equal, more Mp will usually cost you more money. So the question of what size sensor to choose is usually pretty important.

My older (ancient for a digital camera) Epson 750z has 1280 sensors horizontally and 960 sensors vertically. It is a 1.2 Mp camera. My current camera has 2048 sensors horizontally and 1536 sensors vertically for a 3.1 Mp sensor size. It may seem very impressive that the new camera has about three times as many pixels at the older camera. But take a look at the horizontal resolution. The newer camera has 2048 sensors while the older one has 1280. The new camera doesn't even have twice the linear resolution of the older one. In fact, it only has 60% more linear resolution. In order to double the linear resolution, the number of image sensors must be quadrupled.

A 4 Mp camera will have about twice the linear resolution of a 1 Mp camera. And it will take 16 Mp to double the resolution of a 4 Mp camera. As the pixel count increases, 1 Mp steps quickly become somewhat insignificant.

Enlargement

More Mp means more resolution and a greater ability to print a larger image. A lot of the charts seen at retailers are more misinformation for marketing purposes than reasonable guidelines. After we get done here, you can judge for yourself how important the extra Mp will be for you.

It is generally accepted that most people can barely see fine detail at a resolution of about 6 line pairs per millimetre from a distance of about 30 cm. This works out to be about 150 line pairs per inch or 300 image pixels per inch (it takes a minimum of two lines of pixels to resolve a line pair). It is no surprise then that modern photographic printers known for high quality output have resolutions rated at or near 300 ppi (pixels per inch).

In the tests that determine visual acuity, these line pairs are very thin black and white lines of very high contrast. But at the limit of vision, these lines are just barely distinguishable from a solid strip of grey. Detail begins to become indistinct as we get close to the limits of our vision. The point where we lose the ability to make the distinction is where the detail seems to merge together in a blur.

Figure 1. The linear resolution difference between
cameras with different megapixel sizes isn't as great
as the numbers might suggest. It takes four times
the pixels to double the linear resolution.

A minimum of two camera sensors (or pixels) is needed to resolve one line pair. The Nikon Coolpix 4500 is a popular 4 Mp camera with 2272 horizontal sensors. If its lens is up to the task (which it probably is), we'd expect it to be able to resolve 1136 total line pairs. Since the limit of human vision is around 6 lp/mm, this camera should be able to create an image 189 mm long that is as sharp as the eye can see (from a distance of 30 cm). In other words, it should be able to make an impeccable 5x7 print. Few people will be able to see any more detail than this without using some sort of magnifying aid.

This is the degree of resolution that might be desirable for fine art imaging or very demanding print publications. But the reality is that most of us aren't quite that picky about sharpness. It is also true that we don't typically make pictures of very high contrast line pairs. Our subjects usually have much less contrast and usually have small details that are not nearly so regular in their shape or pattern. Some subjects are more likely to reveal the lack of sharpness than others.
Human visual acuity and apparent sharpness is also a more complex issue than can be explained simply by the ability to resolve high contrast line pairs. Factors such as image noise and contrast come into play. We can quite reasonably lower our standards to 240 pixels per inch (4.7 lp/mm) and make prints that most people would consider sharp. This is especially true of images from digital cameras since they tend to produce images that have very low noise and no film grain. It's good that digital cameras have that edge, since the next thing I'm going to tell you is that just about every digital camera actually has a lower real resolution than its Megapixels would lead you to believe.

Masking and Interpolation

Some cameras will enlarge an image and increase the number of Mp through digital interpolation. My Epson 750z does this to increase the image size from 1.2 Mp to almost 2 Mp. If you have a computer and an image processing program, you can do the same thing without using the camera. This same basic process is used with digital zooming. A small section of the CCD is simply resized to the same number of pixels that would be used for a full frame. Neither of these processes can add any more detail than was recorded on the camera's CCD. So while you have a greater pixel count, you don't have any more detail.

Cameras that interpolate to higher Mp sizes have made it necessary for manufacturers to distinguish the differences between Mp ratings. Many manufacturers will now make the point that their Mp ratings reflect the "true" resolution of the CCD sensor. But the truth is that the effective resolution of almost all CCDs in modern cameras is about 25-30 percent less than what the camera's Mp rating would suggest. How can this be?

CCDs and most CMOS sensors used in digital cameras are "colour blind". The individual sensors detect a wide range of light frequencies (colours), but they do not distinguish between these different colours. Each sensor site simply reports back the amount of light that it received, with no information at all about colour. In order to give the sensors the ability to record colour, a process called masking is used. In a masked CCD, each sensor is overlaid with a small red, green or blue filter (sometimes a cyan, magenta, yellow and green filter). A 2x2 sensor site grid typically will have one red sensor site filtered for red, one for blue and two for green. This is called a Bayer mask. So you can easily see that a 4 Mp camera only has 1 Mp of red sensors, 1 Mp of blue sensors and 2 Mp of green sensors. More green sensors are used because human vision is more sensitive to green light. (see Figure 2)

But if you look at a digital pixel in your image editing program, you can see that each pixel has a red, blue and green component. This is accomplished by interpolation.

Figure 2. The colour filter array (Bayer mask) is needed
so that standard sensors can render a colour image.
The actual light sensitive receptor sites are sensitive to
a wide range of wavelengths. Without the filter array,
a sensor could only produce a grayscale image. Each
sensor detects red, green or blue light only.

The camera uses the colour information from adjacent pixels to determine what additional colours should be added to each monochromatic pixel. While it may use a more complex algorithm, the basic idea is that the 1 Mp of red pixels gets interpolated upwards to 4 Mp. The same goes for blue. Green gets interpolated from 2 Mp up to 4 Mp. If you have a program like Adobe Photoshop that enables you to look at the three colour channels independently, you can see that the green layer is almost always sharper and less noisy than the red and blue layers.

The net result is that you get colour information for pixels in all three colour channels, but all of that information has been digitally "zoomed" a bit. The real ability of sensors like this to resolve detail is probably somewhere between 25 and 50 percent less than what the camera's Mp rating might lead you to believe. My personal resolution tests lead me to estimate that it is around 30 percent for a Coolpix 995. I would expect that would be typical for most digital cameras.

Digital SLRs and the Foveon Chip

There is one sensor currently used in a digital camera that does not use masking. This exception is the Foveon X3 sensor. This is very new technology and there is only one camera, the Sigma S9, currently available that uses it. This camera is a 3.5 Mp camera with a 2268x1512 sensor. That might seem unimpressive and very average, but real world testing shows that is has an effective resolution slightly less than a 6 Mp camera that uses a conventional sensor (3000x2000). If we multiply 3000 pixels by 70 percent (to account for the approximately 30 percent reduction in resolution that we'd expect from the colour filter mask) we end up with 2100 pixels. That is right in the ballpark of what we would expect to see. Virtually all current digital cameras use masking and consequently their true effective resolution is somewhat less than what their Megapixel rating would suggest.

While I'm on the subject of the Sigma S9, I should mention that there is a significant difference between the sensors typically used in expensive digital SLR cameras and those used in consumer or prosumer digital cameras. Digital SLR cameras typically use sensors that are much larger with individual sensor sites that are also larger than the ones found in the non-SLR other cameras. It seems that larger sensors tend to produce less noise and can have greater light sensitivity. It seems that the "sweet spot" for individual sensor size is 5-9 µm. The non-SLR consumer cameras typically have the smaller CCDs (8-11 mm diagonally) and smaller sensor sites as well (3.5 µm and even smaller). This is yet another way that all Megapixels are not created equal.

True Resolution

So, our example Coolpix 4500 with its 2272 pixels probably won't give us 1136 line pairs horizontally. It will probably yield closer to 800 line pairs. If we print at 8x10, we can expect an image that delivers a little more than 3 lp/mm. This doesn't sound particularly impressive when compared to the standard of 6 lp/mm - and really it isn't. But let's remember that our standard is based on the idea 6 lp/mm is usually the limit of our ability to resolve detail and that most people will consider a 4.7 lp/mm photograph to be sharp (240 pixels per inch). It is also true that digital cameras typically have no film grain and overall less noise than film cameras. This gives a digital camera an advantage in perceived image quality. Especially when large areas of low detail (such as the sky) are featured in an image.

Modulation Transfer Function

I have been using resolution as a way of estimating picture quality and the degree of enlargement that you can expect from a digital camera. I've also mentioned the positive effects of low noise on perceived image quality. These are important factors, but human visual perception is more complicated still. The contrast in the rendered detail plays a very large part in what we perceive as sharp. This can be measured as the imaging system's Modulation Transfer Function (MTF). This measure gives information about contrast at specific levels of detail. It essentially tells us not only what level of detail was resolved, it tells how accurately it was resolved at a particular level of detail. A series of black and white bars is typically used much like those used in a resolution test. As the level of detail increases, the accuracy (contrast) tends to diminish. Eventually we get gray mush (zero percent contrast) at the limit of resolution. The point where contrast is 50 percent accurate is a good indicator of perceived sharpness (resolution at 50 percent MTF).

Figure 3. The Foveon® sensor actually uses three layers
of sensors and takes advantage of the fact that the different wavelengths of light penetrate silicon to different depths.
This sensor gathers red, green and blue light at each pixel
location.

The judicious application of sharpening tools in programs such as Photoshop, Paint Shop Pro and others can usually improve the MTF response so that this 50 percent point is reached at a higher resolution. Sharpening can't add detail or improve resolution, but it can alter the perception or impression of detail. Using such programs is quite easy with images from a digital camera.

Putting It All Together

So, when all is said and done, a good 4 Mp camera can make a pretty decent looking 8x10 picture that might fall slightly short of the standards for a quality publication or fine art. A 2 or 3 Mp camera won't be far behind and would probably be able to make 8x10 prints that please most people. I have been using the 8x10 print as a standard of reference throughout this article. You might conclude then, that it would be pointless to make an 11x14 or larger print from a 4 Mp camera. That would be incorrect. The 8x10 standard is based on a viewing distance of 30 cm (about 12 inches). The greater the viewing distance the more an image can reasonably be enlarged. At 60 cm, the picture only needs to be half as sharp. Very nice images larger than 11x14 (I've printed up to 12x16 inches) can be created as long as the viewing distance is appropriately increased. Just don't expect the image to hold up well as you get your nose closer to the print. I have a very nice image of a Greater Yellowlegs in our bathroom that was printed at 12x16 from a 3 Mp camera. You would have to stand on plumbing in order to get very close to it. It looks quite nice at the viewing distance I have imposed on the viewer.

It would be easy to pick some standard, do a bit of multiplication and then crank out a chart equating Mp to image enlargement. But such a chart would be too simplistic. If you want to make such a chart, you now have some information that you base it on. But how many line pairs per millimetre are enough? The answer is pretty clear for top quality publications and most professional work. For other applications, it is much less obvious. My suggestion is that you see for yourself. There are a number of Web sites that provide full resolution, unaltered test images taken with a wide variety of cameras. You can download these images at no cost and have them printed as you expect to have your own images printed. You can then judge for yourself how many Megapixels will be enough.

About the Author
Jay Turberville, jay@studio522.com, is a graphics designer, animator, editor and cofounder of Studio 522 Productions, Inc. in Phoenix, AZ USA www.studio522.com. Jay is an amateur photographer currently putting most of his photography energies into shooting images of birds using telescopes and digital cameras ( digiscoping).http://www.jayandwanda.com.

Reprinted from the March 2003 issue of PC Update, the magazine of Melbourne PC User Group, Australia

[ About Melbourne PC User Group ]