Sad day - last of the single digit sensors?

Started Apr 7, 2009 | Discussions thread
Veteran MemberPosts: 6,274
Re: A misconception .. by DPReview???
In reply to eNo, Apr 13, 2009

eNo wrote:

Yeah, still waiting for the eye-popping explanation of how poorer
noise performance (lower SNR) on each photosite collector, when
integrated with all other componet interplays and aggregated with
pixie dust leads to improved or equivalent signal quality.

SNR is a scale dependent quantity, so yes, photosite scale S/N decreases with decreasing photosite size. But this tells you nothing about whether the S/N of the image is going up or down with photosite size, since comparing two photosite scale S/N levels of cameras with different photosite sizes is comparing two functions (the S/N as a function of spatial frequency) at different values of their arguments; a meaningful comparison of the value of two functions should at least be made at the same scale.

Therefore one should do the proper scaling; the analysis can be found at for instance

And if theory is not your thing, there is always the corresponding experiment:

Blue is Canon 50D native, Red is Canon 40D native; Orange is 50D resampled with PS Bicubic, Black is 50D resampled with ImageMagick Lanczos. On this graph, 256 on the horizontal axis is Nyquist of the 50D; 209 is Nyquist of the 40D. We see that the noise power is more or less the same for all images with the pixel dimensions of the 40D (with Lanczos being slightly better than PS Bicubic), and that all that has been done by resampling is to chop off the spatial frequency response at higher frequencies beyond Nyquist of the 40D (as expected). This not only eliminates all the detail due to higher spatial frequencies, it also eliminates the noise components arising from those frequencies, which is why the std dev of the histogram decreases -- the std dev is proportional to the sqrt of the area under the curve.

If you want to look at the images on which the analysis is based, they are linked to here:

The point is that these two cameras of rather different pixel density have comparable S/N when it is measured at the same spatial frequency scale.

Downsampling, which is often used as an method of comparison, is an inessential ingredient of the analysis -- downsampling does not affect the noise power appreciably below the spatial frequency cutoff introduced by the downsampling, as the above graph shows. It is merely sometimes useful as a means of visual comparison on computer screens, which have a fixed scale that the image is being mapped to.

Pixel scale std dev, often used as a method of comparison, is not helpful because it is an average over spatial frequencies, which includes the noisier high frequencies when a camera has a smaller pixel pitch. For instance, a smooth patch in the images analyzed yielded std devs of

40D: 3.23
50D: 4.28
50D dowsampled: 3.08

and yet the 50D image and its downsampled counterpart have the same noise frequency spectrum over the range of frequencies that they have in common.

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