I've always ignored this as probably irrelevant for my purposes but I would like to know what the difference between "Total Pixels" and maximum image size pixels is. For example the new Nikon D4 has 16.6 total megapixels but it has 16.2 megapixels at the 4928 x 3280 maximum image size. Can anyone clear this up for me?

There simply are more pixels on the sensor than the camera actually uses. maybe another row and column. So the actual imaging sensor as produced is 16.6 but only 16.2 are used for the final image. All cameras do this.

Thanks, E.J.,

That's what I figured but why claim 16.6 total pixels? It seems rather meaningless. Why not use the maximum image size, which is relevant?

It is completely meaningless and simply marketing.

There are a few rows and columns on the edge of the frame; some are masked off from incident light. If there were more of them, they would be statistically useful in subtracting line noise (banding) from the image; unfortunately there are not enough. It is not clear to me whether there is some use that the manufacturers use them for in firmware or proprietary software.

the extra pixels are a reference to correct for thermal drift. They are masked permanently. The banding noise comes from readout circuit and cross-talk. It cannot be corrected by using reference pixels.

The manufacturer states the total number of pixels that is fabricated in the sensor, just like a chip manufacturer like Intel states the total number of transistors on a chip, not all of them are used for "effective" data processing...I don't see a problem with that. Just like you buy a hard disk the usable capacity is lower than the advertised capacity.

My understanding is that in addition to the masked pixels mentioned above, there are also some extra pixels around the edges of the sensor that are exposed to light but the information gathered is only used by the demosaicing algorithm to determine the color for the "effective pixels" at the edges of the resulting image.

humm...I don't understand what you mean by "color of effective pixel at the edge"...do you have a reference?

There was an article about this on Luminous Landscape some time ago. He had a link to a program that would recover the extra pixels from the raw file. I believe the gist of it was that the border pixels(aka ones that we see)have their color values determined by surround pixels, hence the need for "extra" pixels around the actual image pixels.

I believe he's saying that since it takes several surrounding sensor pixels to determine the final color of an image pixel, the extra sensor pixels could be used to determine the final color of pixels at the edge of the image. Assuming the demosaic algorithm uses an array of sensor pixels to determine the color on 1 image pixel.

I have no idea if that's what actually happens.

David and Scott answered (thanks!) while I was trying to compose a reply and their interpretations are correct.
Here is my more long-winded (although not necessarily better) attempt to explain what I meant :

To simplify the explanation, let's forget about the masked pixels for a minute. In general, each pixel (photosite) on the sensor maps one to one with a pixel in the final image file. But since each photosite only sees one color of light, the demosaicing algorithm must also look at surrounding photosites to compute a full complement RGB value. If the number of photosites on the sensor (total pixels) was equal to the number of pixels in the resulting image file (effective pixels), there would not be enough information to come up with fully accurate color for the pixels at the edges of the image file because there are not enough surrounding photosites for the demosaicing algorithm to use (the photosites at the edge would only be surrounded on one side). Therefore, extra photosites are added that capture light but are only used in order to compute the RGB values for the pixels at the edges of the final image file, i.e. these extra photosites do not themselves map directly to any pixels in the final image.

Here are 2 places where this is discussed :

http://www.dpreview.com/learn/?/Glossary/Camera_System/effective_pixels_01.htm

http://photo.stackexchange.com/questions/9738/why-are-effective-pixels-greater-than-the-actual-resolution

The above explanation is based on my attempts to understand this stuff at a conceptual level after reading the above information. Take a look and see if you agree.

Note that if the above were true, there would have to be more photosite values in the RAW file than there are RGB pixels in the image after RAW conversion. Is there any way to check this?

If you do not mind compiling code to take a peek at the extra data, take a look at dcraw code
http://www.cybercom.net/~dcoffin/dcraw/

Also, simplest explanation on "why" can also be found here.

Why are dcraw output images larger than camera JPEGs?
Any algorithm that combines each pixel with its neighbors is going to have problems near the edges. C code is cheap, so dcraw applies a different algorithm to edge pixels. Hardware logic is expensive, so cameras crop off the edge pixels after processing.

Thanks! I figured what you were referring to and it is correct. You just need a single-pixel border on each side to deal with that so for example in a X*Y pixel sensor the number of outer border pixels that you need is 2*(X+Y+2) so for a 16.2 Mpixel sensor (4928 X 3280) the number of outer border pixels are ~15K pixel. this is small compared to the extra 400K pixels that serve as row/column references

I use dcraw a lot, especially with my E-PL1, and so I see there are a bunch of extra pixels on the edges, not just one row/column of them. The extra pixels actually saved me from a compositional problem in a shot I took in Osaka a couple of months ago, when I was in a hurry. I suspect the extra pixels are used by the camera firmware to correct barrel distortion when it occurs. I do notice with the E-PL1's kit lens that the raw files have more distortion than the jpeg's at wide angle, so the firmware is doing something to correct distortion.