Estimating the noise in radio images about fidelity HOT 3 OPEN

I think strong positives pixels (these pixels are not part of a giving source as generally a source is not defined by a single pixel) contribute also, It might be good if
a threshold is giving for positives pixels, if any pixel value is above the
threshold then this pixel highly contribute in artefacts (a sort of
sidelobes contribution from sources that are poorly deconvolve).

On 2 April 2015 at 22:09, Sphesihle Makhathini [email protected]
wrote:

I've been thinking of how best to estimate the noise in an image, because
simply taking the std dev or mad (median absolute dev) of the pixels does
not give accurate results in the presence of bright emission. But, since
the sky emission is positive, the negative pixels should give a far better
estimate of the noise. This is what I'm thinking:

Let D be a N_x_N image and d be the negative pixels of D (1x_M_,
where M is the number of negative pixels).
Then, assuming that the noise is symmetric about zero the noise can be
approximated as:

sigma = std_dev[d || (-d)], where || represents concatenation.

In the presence of calibration artefacts (which generally have negative
counterparts) this might be a bit off, but it should still be an
improvement over using all the pixels.

—
Reply to this email directly or view it on GitHub
#4.

from fidelity.

Hey guys, I am just getting back to thinking about this now.

This is a fine idea, worth trying out. The PSF will have negative
components (very dependent on the weighting), but the fact that we know the
sky is positive we know that any region with an abnormally high number of
negative pixels is a good indication of an artefact. You can probably
improve on this idea by subtracting off any local average due to large
scale structure(only really important for short baseline arrays, and it
will have no effect on long baseline because the average will be nearly 0
already).

There is another step here about looking for local structure by performing
a correlation between all pixel pairs within a limited region.

On Fri, Apr 3, 2015 at 11:57 AM, atemkeng [email protected] wrote:

I think strong positives pixels contribute also, It might be good if
a threshold is giving for positives pixels, if any pixel value is above the
threshold then this pixel highly contribute in artefacts (a sort of
sidelobes contribution from sources that are poorly deconvolve).

On 2 April 2015 at 22:09, Sphesihle Makhathini [email protected]
wrote:

I've been thinking of how best to estimate the noise in an image, because
simply taking the std dev or mad (median absolute dev) of the pixels does
not give accurate results in the presence of bright emission. But, since
the sky emission is positive, the negative pixels should give a far
better
estimate of the noise. This is what I'm thinking:

Let D be a N_x_N image and d be the negative pixels of D (1x_M_,
where M is the number of negative pixels).
Then, assuming that the noise is symmetric about zero the noise can be
approximated as:

sigma = std_dev[d || (-d)], where || represents concatenation.

In the presence of calibration artefacts (which generally have negative
counterparts) this might be a bit off, but it should still be an
improvement over using all the pixels.

—
Reply to this email directly or view it on GitHub
#4.

—
Reply to this email directly or view it on GitHub
#4 (comment).

from fidelity.

I like the idea of subtracting the local average. And correlating neighbouring pixels is actually a good way to find local structure, we should implement this.

from fidelity.