image with noise and sensor grid?

[rubecula]

About 2/3 weeks ago I got a set of Ha images of the Wizard - all were ok and they stacked and stretched with no issues.  The other night during an unexpected clear spell I managed to get 24 x 600sec OIII images.  A jpeg of the stacked and stretched images is below.  Except for the top corners it shows an unacceptable amount of noise and there is a feint grid visible.  Is this from the sensor?  An extreme stretch of the Ha data didn't show the grid.  I thought it might be condensation but on both occasions I was using a dew shield and the dew heaters were on.

The only thing that changed is that there was a Windows update between sessions.  The Atik 383L+ camera and the filter wheel wouldn't connect so I re-installed the driver and everything connected OK.  I've experienced that before and had to re-install just the Atik driver after a Windows update. But in previous cases the resulting images have been fine.

Anyway does anyone have any idea as to what is causing the noise and feint grid at the bottom in the OIII?

Thanks,

Edited October 21, 2019 by rubecula

[tooth_dr]

Hi Robin

It's hard to make out the grid on my monitor here.  I'm noticing that it is out of focus.  Could this be creating problems in the image?

[rubecula]

Yeah it is out of focus.  The numbers in SGP looked ok so I thought I'd play with the image as I had nothing else to work on :-).

The "grid" is definitely there particularly in the bottom third of the image and the horizontals stand out a bit more.

If the grid is from the sensor, and I'm not sure either way, then I wouldn't have thought being out of focus would cause the grid and excessive noise.  But what do I know?

[vlaiv]

Not sure if it has something to do with Windows update.

First let address issue of noise everywhere but top part of the image. Noise that you are seeing is background noise consisting out of read noise, dark current noise (all across the frame) and LP noise. LP signal often has gradient to it because sky is not uniformly lit up by ground light sources - closer to horizon there is more LP and near the zenith there is less - that creates gradient.

You can see that there is gradient running from top of the image down to bottom - that is most likely LP gradient. Your level of stretch is such that it managed to push top part of the image (or rather it's background) - to be dark enough so that noise does not stand out as much while bottom of the image is still bright enough for noise to show.

Try removing gradient first and doing less aggressive histogram stretch so that background noise is better controlled.

Now let's look at pattern in the image. My guess is that it is not related to the sensor at all, and it is due to noise visible and another effect. You are probably guiding but your polar alignment is not spot on. This creates slight rotation between subs. Slight rotation between subs can happen if you have cone / orthogonality error - once you flip meridian frame will be slightly rotated. If frames gradually change rotation - it's due to PA. If one side of pier has one orientation, and other side of pier after meridian flip is slightly rotated - it is due to cone / orthogonality error (I'm not expert in that field, still don't know why exactly it happens, but I know it does sometimes).

Now let's see what that has to do with grid pattern? When you start stacking your images - subs need to be aligned, so some of them need to be "rotated back" to match reference frame. Software uses certain interpolating algorithm to do this, and choice of algorithm can impact creation of the grid pattern. Particularly bilinear resampling can cause that. Here is synthetic example done on just pure noise and stretched to show grid forming:

This is pure gaussian noise (otherwise no pattern is visible), rotated by 1 degree using bilinear resampling method, and stretched to show grid clearly. Depending on rotation angle, grid will be coarser or finer.

That is what I believe is going on, but might be wrong. If you want to lessen this effect- use bicubic resampling or some more advanced resampling technique (lanczos for example is excellent).

[rubecula]

Wow vlaiv! What a wonderful and comprehensive answer.  You've given me so much to do and think about.

Thank you so much