When Is a Megapixel Not a Megapixel

...and how many MP does your camera really have?

Back to the Nitty-Gritty section,
with articles on technical aspects of photography.

One of the most common questions I'm receiving in my daily email load are these concerning pixel count and various file sizes which can be generated by the same camera. Just yesterday (as of the original writing) I received six related emails (well, three from the same person, but still).

I've been planning for some time to write on this subject. Other things always pop up, being more urgent, so the task is being hopelessly delayed. This is why I decided to write just this small technical note. At some moment it will evolve into a full article, for the time being, I hope, it still may help to satisfy your hunger for information.

What really is a pixel?

Strictly speaking, the camera sensor does not have pixels (as an RGB pixel is a point with information on all three color components). It has a given number of photosites, which actually are monochrome pixels (i.e. providing just one, not three values). Each photosite has a tiny R(ed), G(reen), or B(lue) filter on top, which means that it carries information on just one component of the image at that point. Therefore, to be exact, a "ten-megapixel" camera (note the quotes) actually collects about 10/3 = 3.3 megapixels of information, with each layer somewhat shifted from the others.

I'm simplifying things a bit, as in most cameras 50% of photosites are G, 25% R, and 25% B, but this just shifts my numbers a little. I'm also putting aside the question of color depth of the recorded information, i.e., number of bits per photosite, which defines how many different light values per each photosite can your camera capture.

After a picture is taken, a translation process is applied to the original photosite information, translating it into pixels, i.e., points with full RGB information. This is usually done before the image is saved to a memory card, unless you're using the raw file format, in which case it is postponed to the stage of off-camera raw development.

A ten-megapixel camera using the "native" resolution creates a 10 MP file, placing a pixel at each photosite location (or rather "logical location" in the file). In our simplified model, to create an RGB pixel where, say, an R photosite is, the camera will use the actual R information from that place, while recreating the approximate values of G and B by interpolation and averaging (smoothing) from the appropriate neighbors.

This means that regardless of resolution, the captured image is always interpolated, period, and sensors are providing just one-third of the pixel count used to store images. You thought you had a ten-megapixel camera? How does "three megapixels" sound?

In-camera resolution downsizing

Having the built-in math to do the interpolation, the camera does not have to place your logical pixels where the original photosites are. If your 8 MP camera is saving your images as 2MP files, each of those two million pixels will be interpolated from the nearby original photosites as needed. Actually, you may have more pixels in file than photosites in the camera. This is what Olympus does in the C-5060/7070WZ in their "ENLARGE SIZE" mode (in addition to slaughtering the English language). This is also what Fuji did in a number of models, where the pixel grid not only had more points than the photosite grid, but also was rotated at 45 degrees with respect to the latter.

In a good implementation (and I believe all major makers use that), the in-camera image resizing (i.e., when saving, say, 2MP files) still uses the information from all photosites, but converts it into smaller number of pixels.

While this leads to some information loss (obviously!), the loss is less than proportional to the reduction in pixel count, as each pixel uses now information of more photosites (I'm simplifying again for the sake of argument), so it is more "solid" — read: less noisy.

Reducing the file size in postprocessing brings similar results, with a slight difference: in this case you do one interpolation in-camera (filling the gaps for particular color component), and another one in the postprocessing program. Theoretically, one interpolation is better than two, so 2MP images straight from the camera may be slightly better than those reduced in postprocessing, but I think the difference is rather negligible. Besides, in postprocessing you have much more computing power, so you may use smarter (and more CPU/RAM intensive algorithms).

If you have a 5MP camera and you are sure you'll not be making prints bigger than 8x10", there is nothing wrong with using the three-MP mode. If all you care about are just vacation snapshots, you may even go down to 2MP. You will gain a lot in terms of images you may save on a card, and your 2MP pictures from a 5MP camera will be better than those from a 2MP cameras in the "native" mode.

On the other hand, I'm usually not recommending this. Memory cards are cheap, and you never know if some pictures will not deserve a bigger print one day. Still, my wife uses her 6-MP Olympus C-60 saving images as 4-MP files. No harm done: there is still plenty of pixel count left, the lens (while satisfactory) is not really capable of matching that count with optical resolution, and image files are a third smaller, so why not?

Sigma SA-series: three or ten megapixels?

The only cameras which use 1:1 pixel-to-photosite ratio are the SA-series SLRs made by Sigma. While they are using a 3MP grid, each node has three photosites placed on top of each other. This is 9 million photosites, and if the color layers were shifted, the market would recognize the Sigma as a 9MP camera. In reality, the amount of information captured in both cases is identical. Taking a Sigma image into your postprocessing program and enlarging it to 9MP should provide images of quality comparable to "nine-megapixel" cameras (quotes intentional!).

At any moment I would choose the "three-megapixel" Sigma SA-10 SLR over any sub-$500 "eight megapixel" model from any other maker. Well, Sigma has alreadyy switched from three to ten megapixels in their promotional materials, without changing the sensor; what an inexpensive way to increase the pixel count! On the other hand, this levels the playing field, so I don't think it is a misrepresentation.

The megapixel rat race

Unfortunately, the market is specs-driven, and people are used to "larger number means better" paradigm. An Athlon XP processor at 2.2GHz runs at least as fast as a Pentium 4 at 3GHz, and much faster than a Celeron at the same clock speed, and who cares? The usual question is: "How many gigahertz your computer runs at?"

Pixel count might have been a limiting factor of image quality in the late 90's, when cameras were evolving from 0.3 MP to 1 MP and beyond. This is, however, history, except that the mass market never became aware of the fact.

Once you reach four or five megapixels of chip resolution (or rather that count of photosites), other factors are becoming more important than pixel count. That's why some professional-class, expensive cameras claim "only" five MP of resolution, while some very simple ones, aimed at an ignorant audience, sport 6MP or more.

Increasing the pixel count may actually be the cheapest way to improve the on-paper specs of a camera, less expensive than better optics, or even better color handling (not limited to providing more bits per color).

The mass market is, however, largely ignorant, and manufacturers are using that to their advantage.

Next time someone asks you how many megapixels does your camera have, tell them to read this article.

Back to the Nitty-Gritty section,
with articles on technical aspects of photography.

Home: wrotniak.net | Search this site | Change font size

Photo News | The Gallery

Posted 2004/11/14; last updated 2007/08/04; cleaned up 2013/11/18 Copyright © 2004-2006 by J. Andrzej Wrotniak