Silly question

[Admiral]

Ok, on what base (how) do you calculate the pixel resolution of a digital cam ?

[LvR]

Specs of the ccd - ie - the number of pixels found on the ccd...............

Wanna be a bit more specific on the question?

[az]

You just multiply image height by image width (in pixels) and then round off to the nearest hundred thousand. Cameras often have a Total and an Effective Megapixel rating, the only one that matters are effective pixels. The rest are not recorded (It's like the space on a CRT that is blacked out, kind of.). Interpolated values are worthless, of course, only "real", physical pixels count.

Now sensor SIZE or diameter in inches, that's hard, because it is based on some stupid ancient tradition of cathode ray tube sizing. A 2/3" sensor is not 2/3 of an inch in diameter, it is quite a lot less. Details can be found here.

AZ

[mmp121]

Re: Silly question

Originally posted by Admiral
Ok, on what base (how) do you calculate the pixel resolution of a digital cam ?

Base 10 ?

Or do you mean 1000 vs 1024 ?

or do you mean 1280 x 1024 resolution -> 1310720 which is 1.3MP ?

[Admiral]

Say you can do 1600x1200 max picture resolution and the camera is rated for 4mp. How do you get to the 4mp rating ?

[UtwigMU]

or for instance 2272x1704px = 3.871.488px =~ 4mpx.

[LvR]

Say you can do 1600x1200 max picture resolution and the camera is rated for 4mp. How do you get to the 4mp rating ?

If thats a real-world spec by some manufacturer on one of his products he is talking BS.

1600x1200 as AZ said is the spec of a camera with a 2M pixel ccd - saying that camera is a 4M item in the specs implies the manufacturer relies on interpolation (Fuji likes to do this) to achieve the 4M pixels in the file-size - the cam can in reality not produce anything better than 2M rez

[spadnos]

To make things even worse, most cameras are rated for the number of SUB-pixels they have. They've already done some interpolation by combining RGB triplets (or RGGB quads) into pixels.
You can think of it as moving a 2x2 subpixel window over the entire sensor - each pixel is the sum of the single R, single B, and two G subpixels seen through the 2x2 window. So each subpixel is used in 4 adjacent pixels.

- Steve

[Admiral]

Originally posted by LvR
If thats a real-world spec by some manufacturer on one of his products he is talking BS.

Nah, that was BS from me, Utwig and Az got it right. So it's based on the maximum resolution and you just multiply height and width.
Told you it was a silly question

[VJ]

Originally posted by spadnos
To make things even worse, most cameras are rated for the number of SUB-pixels they have. They've already done some interpolation by combining RGB triplets (or RGGB quads) into pixels.

And to make things even more worse (if that is possible), there is a sensormanufacturer which has the subpixels on the same position. Foveon (used in Sigma and some Panasonic cameras) has the 3 subpixels (red, green, blue) on the exact same position). This of course (theoretically) yields great colour reproduction and sharpness, but it makes their megapixel specification a bit weird. You then get 1280x1024x3 = 1.3 megapixel x3 = 4 megapixel... (ah yes, because the traditional system also counts the number of subpixels ) But pixels are only recorded on 1.3 million locations!

http://www.foveon.com/X3_tech.html

Jörg

[spadnos]

The Foveon sensor is great.

They are wackos for rating their sensor at 3x the resolution, though. (10 million sensor sites!)

- Steve

[Marshmallowman]

my kodak does 1800x1200, but they reasonably call it a 2.1 mp camera..

[az]

Originally posted by spadnos
To make things even worse, most cameras are rated for the number of SUB-pixels they have. They've already done some interpolation by combining RGB triplets (or RGGB quads) into pixels.
You can think of it as moving a 2x2 subpixel window over the entire sensor - each pixel is the sum of the single R, single B, and two G subpixels seen through the 2x2 window. So each subpixel is used in 4 adjacent pixels.

- Steve

Sorry, that's not quite correct (or confusingly worded).

All cameras (except those interpolation cheaters) are rated with real pixels, not subpixels. Each pixel in any camera (Foveon sensors aside) is a REAL photosite on the sensor, it is just colorblind. So an array of color filters (usually red, green, blue, green) is placed in front of the sensor, and each photosite can only see a certain color. They are still real pixels, from a luminosity point of view. COLOR gets interpolated between them (and with that, to an extent brightness, too, of course). Each photosite (and the interpolated color information from adjacent photosites) is used for each pixel, so the number of photosites and pixels matches up perfectly. You could say that for "their" color, hpotosites are real pixels, and for the other two colors, they're interpolated.

BTW, a three megapixel Foveon Sensor does actually have 10 million photosites, it's only that three photosites are used for each pixel, instead of 1/3 real and 2/3 interpolated photosites per pixel.

AZ

[az]

I don't think I was any more clear than Spadnos

AZ

[VJ]

Nope, you weren't...

The fact is: an imagingsensor has colourblind photosites, but each photosites gets a filter, so that you obtain something like this:
RGBGRGBGRGBRGB...
GRGBGRGB
BGRGB...
GBGRGB...

(R=Red, G=Green, B=Blue)
Each of these R or B or G are counted to determine the resolution of the camera. But of course, to obain full colour information for each pixel, information from the neighboring pixels is required.
So, when you have for instance a 4 megapixel camera, it physically has 1 million reds, 1 million blue and 2 million green pixels (has to do with colour perception).
And herein lies the discussion: you measure light at 4 million positions, but you interpolate to get the colourinformation. Consequently, the resolution in practise lower than 4 million positions.
(just imagine you could take a picture of a checkerboard black/white so that each square of the checkerboard perfectly matches 1 photosite : the camera won't see a checkerboard !)

The Foveon has a pattern like so:
AAAAAAA
AAAAAA...
AAAAA...
where A stands for all colours (red, green and blue). To obtain colourinformation for a pixel, you don't need to interpolate with the neighboring ones. But foveon currently counts A as 3 pixels, argueing that this is what happens in regular cameras as well.


Jörg