Pixel Density is GENIUS!

Started Jul 13, 2008 | Discussions thread
John Sheehy Forum Pro • Posts: 21,570
Re: This is *not* about different sensor sizes

RRJackson wrote:

simpy wrote:

To recap, this only breaks down if either (1) the smaller pixels
are less efficient or (2) the read noise of the combined smaller
pixels is significantly larger than that of a larger pixel.

Most binning systems I've seen either require 2 or 4 adjacent
photosites for processing. Not that there's no chance of that
changing, but right now as far as I know there's no practical
application in a consumer product that works without a halving or
quartering of the pixel count. So your 10 megapixel compact becomes a
5 megapixel compact for that extra stop of sensitivity. Or a 2 1/2
megapixel compact for two stops. Where just dropping the pixel count
from 10 to 7 will probably get you a stop of extra sensitivity.

I'd love to see a linear correlation between the two where binning
always works out to the same net gain that larger photosites do, but
so far I haven't seen it in practice.

The Bayer CFA makes color binning of RAW data a bit impractical (B&W is easy; you just add four pixels together). You can always bin the full-res RGB image after demosaicing if you want, though, and this has a similar effect. As long as you are not weighing original pixels unequally, binning or downsampling always reduces the new pixel-level noise. It does not reduce full-image noise though. Binning is useful mainly for demonstrative purposes, and its simplicity, as downsampling algorithms tend to introduce wild cards. Downsampling is dodgy with noise on flat backgrounds, because there is no detail for you to see if the algorithm is sharpening it or not (an algorithm may soften or sharpen both noise and desired detail together) The real problem with noise is competition with detail, not the noise itself, per se, so when the image does contain high-frequency detail, you can see both the noise and detail sharpened, and they exist in a ratio, more than as an absolute, in perception.

The real issue is not whether you bin/downsample, but how large you view the image. If our monitors were 100MP, we wouldn't need to bin or downsample our current resolution images to display them, and doing so to reduce noise would not reduce noise, but just lose detail, if we still stretch them to fill the screen.

-- hide signature --

John

Post (hide subjects) Posted by
BJN
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow