Gradient with offset
Gradient with offset
Hi Robin,
Maybe you know what is going on with the following issue. When I'm using a OSC camera that has no ampglow I don't use flats and if I set an offset I get a gradient on both sides of the image about 1/4 to 1/3 of the image on the left and right. The color of one is tinted toward the blue and the other is tinted towards the red. If I use zero offset I get no gradient. This happens on the 294, 2600 and 585.
Any thoughts.
Don
Maybe you know what is going on with the following issue. When I'm using a OSC camera that has no ampglow I don't use flats and if I set an offset I get a gradient on both sides of the image about 1/4 to 1/3 of the image on the left and right. The color of one is tinted toward the blue and the other is tinted towards the red. If I use zero offset I get no gradient. This happens on the 294, 2600 and 585.
Any thoughts.
Don
- admin
- Site Admin
- Posts: 17569
- Joined: Sat Feb 11, 2017 3:52 pm
- Location: Vale of the White Horse, UK
- Contact:
Re: Gradient with offset
Hi Don,
to be honest, not something that I have encountered (or at least not noticed anyway). It would be interesting to see an image showing the effect - maybe that will help understand what is going on.
It may just be down to how stretched the image is - with zero offset, the histogram peak will be further left and will be more expanded in the 'log scaled' live stack histogram, meaning that the same slope as seen on screen would actually be a stronger stretch. That is, the same effect may be there at higher offsets if you get the image stretched enough to bring it up to a visible level.
cheers,
Robin
to be honest, not something that I have encountered (or at least not noticed anyway). It would be interesting to see an image showing the effect - maybe that will help understand what is going on.
It may just be down to how stretched the image is - with zero offset, the histogram peak will be further left and will be more expanded in the 'log scaled' live stack histogram, meaning that the same slope as seen on screen would actually be a stronger stretch. That is, the same effect may be there at higher offsets if you get the image stretched enough to bring it up to a visible level.
cheers,
Robin
Re: Gradient with offset
Hi Robin,
Here are some jpg copies of zero offset and 12offset showing the gradient problem I have.
Regards,
Don
Here are some jpg copies of zero offset and 12offset showing the gradient problem I have.
Regards,
Don
- admin
- Site Admin
- Posts: 17569
- Joined: Sat Feb 11, 2017 3:52 pm
- Location: Vale of the White Horse, UK
- Contact:
Re: Gradient with offset
Hi Don,
yes, I can see the brighter right hand side in the 12 offset values. Trying to track back to the cause, there are three places that this sort of thing can come from
* The light frame images themselves - for instance a light pollution gradient due to the altitude effects or the moon. If this is the case, it will be visible in the raw light frames with a strong stretch applied - if it's visible there then problem solved...
* The flat frames for 12 offset being asymmetric for some reason - have a look at those flat frames compared to the zero offset ones (again a stretch may help) to see if there is any evidence of the 12 offset flat being darker on the right compared to the left (the flat being darker would boost the light frame more)
* Flat over/under correction due to incorrect bias removal - I can see from the settings files that you are not using a dark, which should be fine as long as the flat and the light exactly match in terms of brightness and colour balance settings so that the 'offset' bit at the end of the filename in the flat is representative of the bias level in the light frames. Again, looking at your flat frame file names I can see that the folder names seem to indicate that all of this is set up correctly, but worth checking the capture settings files for the flats to make sure that the actual settings match the folder names
* Needing to use dark frames rather than just hot pixel correction - this depends on how long your light exposures are. The shortcut of not needing dark frames is pretty reliable for short to medium exposures, but as the exposure time in light frames gets longer, there is more thermal noise in the pixel values coming in each light frame. The bias subtraction doesn't remove this, so it will get flat corrected, potentially leading to flat correction not being correct. See the documentation here on best practice for flat frame correction : https://docs.sharpcap.co.uk/4.1/#Gettin ... Correction
cheers,
Robin
yes, I can see the brighter right hand side in the 12 offset values. Trying to track back to the cause, there are three places that this sort of thing can come from
* The light frame images themselves - for instance a light pollution gradient due to the altitude effects or the moon. If this is the case, it will be visible in the raw light frames with a strong stretch applied - if it's visible there then problem solved...
* The flat frames for 12 offset being asymmetric for some reason - have a look at those flat frames compared to the zero offset ones (again a stretch may help) to see if there is any evidence of the 12 offset flat being darker on the right compared to the left (the flat being darker would boost the light frame more)
* Flat over/under correction due to incorrect bias removal - I can see from the settings files that you are not using a dark, which should be fine as long as the flat and the light exactly match in terms of brightness and colour balance settings so that the 'offset' bit at the end of the filename in the flat is representative of the bias level in the light frames. Again, looking at your flat frame file names I can see that the folder names seem to indicate that all of this is set up correctly, but worth checking the capture settings files for the flats to make sure that the actual settings match the folder names
* Needing to use dark frames rather than just hot pixel correction - this depends on how long your light exposures are. The shortcut of not needing dark frames is pretty reliable for short to medium exposures, but as the exposure time in light frames gets longer, there is more thermal noise in the pixel values coming in each light frame. The bias subtraction doesn't remove this, so it will get flat corrected, potentially leading to flat correction not being correct. See the documentation here on best practice for flat frame correction : https://docs.sharpcap.co.uk/4.1/#Gettin ... Correction
cheers,
Robin
Re: Gradient with offset
The OP indicates he is not using flats, which is probably the problem.
Re: Gradient with offset
Each offset image had a matching offset flat. Reference to the flat used is in the settings txt file that accompanied each image.
Don
Don
Re: Gradient with offset
Here are the matching flats used for the images. Both master flats used -bias subtraction
12offset flat:
https://drive.google.com/file/d/1M6DOcg ... sp=sharing
0 offset flat:
https://drive.google.com/file/d/15ZtRUK ... sp=sharing
12offset flat:
https://drive.google.com/file/d/1M6DOcg ... sp=sharing
0 offset flat:
https://drive.google.com/file/d/15ZtRUK ... sp=sharing
- admin
- Site Admin
- Posts: 17569
- Joined: Sat Feb 11, 2017 3:52 pm
- Location: Vale of the White Horse, UK
- Contact:
Re: Gradient with offset
Hi Don,
thanks for sharing - I took a look at those two flats and can see that the brightness peak in both images is offset left and a little upwards from the center. That means that the right hand side (and particularly the bottom right corner) is the dimmest part of the flat image - where the flat is doing the most correction work. That ties in with those areas seeing excess brightness in the stack - if the bias level being used in the flat correction is wrong, the biggest errors in flat correction will be in those parts of the image.
Could you set the camera up with the settings as for the 12 brightness light frames, except set short exposures (~1ms) and cover the camera. Then look at the mean pixel values shown in the histogram readout. Those will give an idea of the bias ADU level for the camera in that configuration. Based on the information in the flat filename, I would expect a value of about 190, but the FITS headers inside the file suggest that from sensor analysis data the value should be about 120. It might actually be worth including a screenshot of the histogram in those settings so that we can see the various channel means/SD values and see if the bias histogram is cleanly separated from the left hand side.
thanks,
Robin
thanks for sharing - I took a look at those two flats and can see that the brightness peak in both images is offset left and a little upwards from the center. That means that the right hand side (and particularly the bottom right corner) is the dimmest part of the flat image - where the flat is doing the most correction work. That ties in with those areas seeing excess brightness in the stack - if the bias level being used in the flat correction is wrong, the biggest errors in flat correction will be in those parts of the image.
Could you set the camera up with the settings as for the 12 brightness light frames, except set short exposures (~1ms) and cover the camera. Then look at the mean pixel values shown in the histogram readout. Those will give an idea of the bias ADU level for the camera in that configuration. Based on the information in the flat filename, I would expect a value of about 190, but the FITS headers inside the file suggest that from sensor analysis data the value should be about 120. It might actually be worth including a screenshot of the histogram in those settings so that we can see the various channel means/SD values and see if the bias histogram is cleanly separated from the left hand side.
thanks,
Robin
- admin
- Site Admin
- Posts: 17569
- Joined: Sat Feb 11, 2017 3:52 pm
- Location: Vale of the White Horse, UK
- Contact:
Re: Gradient with offset
Hi Don,
thanks for posting those screenshots - I think that I have some ideas now...
When you have the colour balance set to 50/50, the average pixel values are about 120 - close to the value in the FITS headers from the sensor analysis data. However when the colour balance settings are adjusted, that changes the average pixel values of the bias frames, making the red and blue values much bigger (particularly red). The simpler approach to flat correction that you are using (flat only, no darks) isn't going to cope with this situation, since it assumes the same bias level for all 3 colours of pixels.
There are a number of ways to get around this...
1) Leave the colour balance controls at 50/50 at all times and fix colour in post-processing (live stacking colour balance, etc) - this is the way I always work.
2) Use dark frames - if these are captured at the same colour balance settings then they will correctly remove the bias levels for all 3 different colour channels. Combining that with either a monochrome flat (or one captured with dark flats rather than bias frames) should then give good results I think
I don't think I'm going to attempt to make a software workaround for this - the behaviour being seen is dependent on the way that the red/blue white balance controls work (multiplying pixel values by a factor). That's not a universal thing - some colour balance controls for certain makes/models work in terms of altered analogue gains for the channels, which would not show this problem (and a correction for this problem would break things on those cameras). In fact, the way that the ZWO controls work is a bit of an issue, since if you set the value to 60, it's not just adding 20% to the part of the signal due to light hitting the sensor, it's also adding 20% to the bias part of the signal. Ideally cameras would work by adding a percentage to the part of the signal above the bias level.
cheers,
Robin
thanks for posting those screenshots - I think that I have some ideas now...
When you have the colour balance set to 50/50, the average pixel values are about 120 - close to the value in the FITS headers from the sensor analysis data. However when the colour balance settings are adjusted, that changes the average pixel values of the bias frames, making the red and blue values much bigger (particularly red). The simpler approach to flat correction that you are using (flat only, no darks) isn't going to cope with this situation, since it assumes the same bias level for all 3 colours of pixels.
There are a number of ways to get around this...
1) Leave the colour balance controls at 50/50 at all times and fix colour in post-processing (live stacking colour balance, etc) - this is the way I always work.
2) Use dark frames - if these are captured at the same colour balance settings then they will correctly remove the bias levels for all 3 different colour channels. Combining that with either a monochrome flat (or one captured with dark flats rather than bias frames) should then give good results I think
I don't think I'm going to attempt to make a software workaround for this - the behaviour being seen is dependent on the way that the red/blue white balance controls work (multiplying pixel values by a factor). That's not a universal thing - some colour balance controls for certain makes/models work in terms of altered analogue gains for the channels, which would not show this problem (and a correction for this problem would break things on those cameras). In fact, the way that the ZWO controls work is a bit of an issue, since if you set the value to 60, it's not just adding 20% to the part of the signal due to light hitting the sensor, it's also adding 20% to the bias part of the signal. Ideally cameras would work by adding a percentage to the part of the signal above the bias level.
cheers,
Robin