Please explain sensor pixel count and output resolution

[Sassi Haham]

I guess this was asked several times before...

Each sensor has its "effective pixel count", but sometimes outputs a different number of sample points.

For example
my canon xl-2 has around 0.4M pixels on the sensor and the camera records less than that (standard NTSC).
Xl-h1 has 1.5M and the camera output them all to reach full HD using pixel aspect ratio.
The new panasonic GH1 has 12M pixels ! But the camera needs only to output around 2M
of sample points (what is the right term?)!

Does the cpu of this camera takes the extra data and downsample it to 1080x1920 sample points?

If so, what happens if I apply digital crop on a camera like this?
I would still have extra data to start with, would the quality loss won't be significant ?

sassi

[Paul Tauger]

Some camcorders subsample (use the average of a number of pixels to determine the output pixel) but, for the most part, the extra sensor count on cameras like the GH1 is to enable higher-res still photography or image stabilization and contributes nothing to video image.

If you crop digitally you will lose resolution.

[Boyd Ostroff]

Here's a good article about the XL2, since you mention it: Canon XL2 CCD Block Overview by Chris Hurd

I think it varies greatly with the camera, and it gets more complicated because many camera's use "pixel shift" where the R-G-B sensors are slightly offset. This allows them to gain more resolution by averaging groups of pixels.

I think you will find more info on specific cameras by visiting the websites of their manufacturers. I know that Sony published info about how this works on their Z1 and V1 cameras, for example.

[David Heath]

Strangely, manufacturers like to quote big numbers ( ;-) ) so specs can need quite a lot of interpretation. You have to be sure you're comparing like for like - especially the difference between 3 colour pixels, and single colour (R,G or B) pixels.

Firstly, are we thinking of a single or three chip camera? If the latter, are the numbers quoted per chip, or the total of the three?

For full raster 1080 video, the recorded dimensions are 1920x1080 - 2 megapixel in round numbers. For a three chip camera to do this justice, it should really have 2 megapixel each of red, green and blue, so 6 megapixel in total of r, g, b.

It's impossible to directly relate this figure to single chip designs for several reasons, but ROUGHLY a figure of about 4 megapixel or more is likely to be in the right ball park. (Single chip sensors have inherently higher luminance than chrominance resolution.)

So is 12 megapixels a good thing? For stills, yes, but a little more complicated for video. It means the size of each pixel is much smaller than it would be for a 4 megapixel single chip design, which may adversely affect the sensitivity. It may also be far more difficult to accurately downconvert to the output resolution

[Sassi Haham]

Thanks,
SLR cameras for video are knew players and because the CMOS is so large and loaded with pixels they hence subsample to get to 2M pixels for 1080p or 1M pixels for 720p.
Now the GH1 has 14M pixels total, 12M effective for photographic use and I think 10.5M pixels effective for 16:9 video. Ten pixels to subsample for each output pixel of 720p.
applying in-camera digital zoom, will lower the available pixels for subsampling.
Sure applying the in-camera digital zoom will lower the output quality but for what amount I can't predict, I think it would be noncritical to a point.
Does my explanation make logic?
Sassi

[David Heath]

Originally Posted by Sassi Haham

Ten pixels to subsample for each output pixel of 720p.

Ah, but the first ten pixels are of EITHER red, green or blue. The output pixels need to encompass all colours. The theory here gets very complicated, and partly down to relative luminance/chrominance resolutions, but the ratio is far from one to one. "Pixel" has taken on different meanings between chip and output.

for what amount I can't predict, I think it would be noncritical to a point.
Does my explanation make logic?

What your explanation misses is the performance of the downconvertor. Broadcasters spend thousands on buying a good hardware downconvertor - what's put into a consumer camera costs vastly less, and unsurprisingly doesn't do as good a job. That's why it's so difficultto predict actual performance from simple numbers.

To get full 1080 video, you can't beat starting with 3 two megapixel chips.

[Jim Andrada]

Actually, I think it's usually more like Green, Green, Red, Blue in most SLR cameras. If I recall correctly they generally have two sites for green and one each for red and blue.

Anyhow, it's all a sophisticated mathematical game of sampling and the number of sites on the chip are not nice neat little squares that fill up the whole chip as most people would suspect. There are gaps between the sites, and sometimes the sites are arranged in more of a honeycomb pattern (Fuji Film) than a rectangular pattern. And they do typically use half the sites for green (which is where most of the luminance info comes from)
The relationship between sensor sites and output resolution is not necessarily quite as much a clean cut one-to-one relationship as most people imagine