ejmartin

Huh? Read noise went from 3 e- for the D7000 to 4 e- for the D800. Not nearly as good. -- emil --...

The Poisson distribution which applies to shot noise probability has a variance equal to the mean expected number. The latter need not be an integer, since it is the mean expected number; when...

What is your method? -- emil -- http://theory.uchicago.edu/~ejm/pix/logos/logo-rhwp.jpg http://theory.uchicago.edu/~ejm/pix/20d/

Cool! I always wondered where dust bunnies came from... now I know -- it's the Great Red Spot of Jupiter. -- emil --...

But it does seem to have a red spot at the top left. -- emil -- http://theory.uchicago.edu/~ejm/pix/logos/logo-rhwp.jpg http://theory.uchicago.edu/~ejm/pix/20d/

How old is the one pictured above ^^^ ? -- emil -- http://theory.uchicago.edu/~ejm/pix/logos/logo-rhwp.jpg http://theory.uchicago.edu/~ejm/pix/20d/

AHD, VCD, LMMSE, AMaZE, LPA-ICI are examples. These are variations on a general theme of using the color difference channels R-G, B-G interpolated in different directions and seeing which...

This is a bit of a gray area -- there are correlations among the channels that assist in inferring an optimal interpolation. Often these correlations are ignored in the converter's demosaic...

Except at ISO 100, you are looking at the DR of the ADC, not the DR of the pixel. -- emil -- http://theory.uchicago.edu/~ejm/pix/logos/logo-rhwp.jpg http://theory.uchicago.edu/~ejm/pix/20d/

NEX7 seems to use a level thinning strategy similar to Nikon lossy compression. There was a thread a few weeks back in the Sony forum where Iliah displayed the compression curve. ...

It's not the 'Bob theory', it is basic physics and simple scaling analysis. I don't believe he said that, I think he said that deep DoF is not the arena where larger sensors have an advantage...

No correlation of noise at different pixel sites is uncorrelated noise. As an example of correlated noise, pattern noise -- pixels along a vertical or horizontal line have correlated noise...

To the extent that the original noise is uncorrelated, and smoothing is local, you can measure the noise spectrum and extrapolate from the low-frequency portion that is largely free of NR...

Most people look at images in position space, not Fourier space. The issue then is whether the LP filter increases SNR on average across locations in position space. Clearly if you have an image...

I think the point may be that the downsample throws away the spectral band with the lowest S\/N; what is left has higher S\/N and so the image looks cleaner. -- emil --...

I believe in image processing this is called 'soft thresholding'. -- emil -- http://theory.uchicago.edu/~ejm/pix/logos/logo-rhwp.jpg http://theory.uchicago.edu/~ejm/pix/20d/

But if the relative proportion of signal to noise decreases at high frequencies, as it does in natural images, then low-pass filtering will have the visual effect of reducing noise. -- emil --...

Is a blue sky, or smooth skin, not signal? I tried to explain the issue here: http://forums.dpreview.com/forums/read.asp?forum=1018&message=39851009 but it seems to have been...

Both. The issue is 'how much' detail there is at a given scale, relative to noise. In typical images, the strength of signal is a decreasing function of spatial frequency (image scale). I...

Here you go, not exactly on point but closely related to the present discussion: http://forums.dpreview.com/forums/read.asp?forum=1019&message=31922793 The people who know the subject find it...