I've been confused about this. Are camera and display resolutions rated in pixes or sub-pixels. Wiki says sub-pixels, so 1920 x 1080 really means 640 x 360 pixels (which doesn't seem like much because you don't see sub-pixels), and the D800 is only 36 Mega-sub-pixels (4Mp)? I was just trying to calculate video bit-rate and comming up short.
pixels vs sub-pixels
- Thread starter ralfmouth
- Start date
It's pixels. Sub pixels aren't a thing unless you're specifically calculating for them, so it would be a totally useless measurement and would be impossible to apply to image recording. If the camera recorded something that can be defined as a point that is, by definition, a pixel. Unless we're talking about different things. Where, exactly, did you read that sub pixels were used for measurement?
Samuel H
Major Contributor
it's a bit more complicated than that... I'd say that for sensors it's actually pixels (but monochrome pixels), whereas for LCD screens it is usually sub-pixels.
* Sensor: when canon says "18 Mpix", they actually mean 18 milion pixels, but each of them only has information about one color (it's either a red, or a green, or a blue pixel). The debayering algorithm interpolates the information so that your final JPEG image can have 18 million full-RGB pixels; but this doesn't generate any new information, it's just interpolation from the original 18 million "monochrome" pixels (with good math, the final image can look pretty neat, as you see in the still images from all modern cameras, so it is this Mpix number that you should worry about)
* LCD screen: when canon says "1040 K dots", they actually mean "around 1040000 sub-pixels", again each one with only a single piece of color information (either R, or G, or B), and in this case also sharing spatial information (the R, G and B sub-pixels are supposed to be on top of each other, whereas on the Bayer pattern of the sensor they sit by one another). So, 1040 Kdots means 720x480 (and in this case this "real resolution" is what you should worry about).
* Sensor: when canon says "18 Mpix", they actually mean 18 milion pixels, but each of them only has information about one color (it's either a red, or a green, or a blue pixel). The debayering algorithm interpolates the information so that your final JPEG image can have 18 million full-RGB pixels; but this doesn't generate any new information, it's just interpolation from the original 18 million "monochrome" pixels (with good math, the final image can look pretty neat, as you see in the still images from all modern cameras, so it is this Mpix number that you should worry about)
* LCD screen: when canon says "1040 K dots", they actually mean "around 1040000 sub-pixels", again each one with only a single piece of color information (either R, or G, or B), and in this case also sharing spatial information (the R, G and B sub-pixels are supposed to be on top of each other, whereas on the Bayer pattern of the sensor they sit by one another). So, 1040 Kdots means 720x480 (and in this case this "real resolution" is what you should worry about).
Here:
"LCDs typically divide each pixel horizontally into three subpixels. In the display industry terminology, subpixels are often referred to as pixels"
Well it just took me 3 hours with a magnifying glass to count every pixel on my HDTV and....
Just kidding, extrapolating from 1/16" gave exactly 1920 horizontal "true" pixels (each with rgb).
So you're right. Not sure what context the above refers to.
Most digital camera image sensors use single-color sensor regions, for example using the Bayer filter pattern, and in the camera industry these are known as pixels just like in the display industry, not subpixels.
So cameras are actually rated in sub-pixels which is what i think you said (a monichrome pixel is a sub-pixel?)
Which means technically, resolution = 1/3 (though interpolation increases perceived resolution)
Technically 1920 x 1080 requires an 18MP camera but i see it's not that simple.
Samuel H
Major Contributor
Well, again... yes and no
The important thing is: as long as you know what's really going on, what words you use to describe it really don't matter much.
On a sensor there are a number of monochrome pixels with a color filter in front of them, so each one is either R, or G, or B, and the final image is usually a complex interpolation of the information from all these monochrome pixels. I would call them pixels and not subpixels, but that's as I said above: as long as you know what's going on, words are just a matter of agreeing on which ones to use (and I think here the common thing is to call them pixels).
Regarding how much information there is, with a 18 Mpix sensor that works as described above, you can create a YUV or YCbCr image that has full 18 Mpix resolution in the luminance channel (i.e. you can extract luminance information from each "monochrome pixel with a color filter in front of it"), but only a quarter of that in the color channels (actually it will have 18mpix of resolution there too, but it will be very soft because it relies too much on interpolation: you only have a "full set of color information" for every 4 pixels). So you can think of this 18mpix Bayer sensor as having a resolution of 18mpix but also 4:2:0 color subsampling "built in".
The important thing is: as long as you know what's really going on, what words you use to describe it really don't matter much.
On a sensor there are a number of monochrome pixels with a color filter in front of them, so each one is either R, or G, or B, and the final image is usually a complex interpolation of the information from all these monochrome pixels. I would call them pixels and not subpixels, but that's as I said above: as long as you know what's going on, words are just a matter of agreeing on which ones to use (and I think here the common thing is to call them pixels).
Regarding how much information there is, with a 18 Mpix sensor that works as described above, you can create a YUV or YCbCr image that has full 18 Mpix resolution in the luminance channel (i.e. you can extract luminance information from each "monochrome pixel with a color filter in front of it"), but only a quarter of that in the color channels (actually it will have 18mpix of resolution there too, but it will be very soft because it relies too much on interpolation: you only have a "full set of color information" for every 4 pixels). So you can think of this 18mpix Bayer sensor as having a resolution of 18mpix but also 4:2:0 color subsampling "built in".
Last edited:
Breaking it into luminance and chroma really clears things up. Thanks!