More megapixels, better photos: Fact or fiction?

Started Feb 7, 2008 | Discussions thread
ejmartin Veteran Member • Posts: 6,274
Re: So many misinformed replies

cpw wrote:

ejmartin wrote:


While the individual parcels have more
variance, since a fixed area of sensor has the same number of photons
falling on it, a fixed area of the image still has the same number of
photons and thus the same sample variance and hence the same noise on
a per area basis. The only thing that more pixels brings is more
resolution, and that's a good thing.


Such a general statement as this, i.e. per area noise equality, only
strictly holds if there is only shot noise. However, each pixel has
read noise on it, it has to taken in to account (I know you talk
about read noise below).

Where in my initial post did I talk about shot noise and read noise? I talked about sample variance and sampling error for large and small samples. That those variances and errors have multiple sources doesn't matter so much to a first approximation.

Suppose you are asked to quickly estimate the size of a crowd of people. Do you think that if the size of the crowd is about 10,000 that you will make the same error in your estimate as when the size of the crowd is about 100? The crowd size isn't fluctuating, one just has the analog of read noise here. So read noise varies with pixel size. Shot noise does too, because it quite directly varies with (square root of) sample size.

Why do you think read noise goes down with pixel size in the examples I listed? It's because whatever is counting the electrons has a smaller sample to deal with, and thus makes a smaller counting error.

To see this, look at an extreme case. Suppose 10,000 e- hits an area
patch. One camera has 1 pixel for this area, and a second camera has
10,000 pixels. Both have 20 e- read noise. The SNR for the area in
the first camera would be 98. The SNR for the second camera would be
about 5, because you have to rss the noise contributions (which in
this case is read noise dominated for the second camera) from all of
those little pixels.

If we're going to talk about hypothetical situations, they should reflect reality closely lest wrong conclusions are drawn. That's why in the further discussion I chose measured data from actual production cameras.

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