Colour gradients during DBE

[peterbrack]

I would like to know if there is any way off preventing, or it not preventing, fixing the colour gradients that  start from DBE.

The image attached is unfinished LRGB but the problem looks like this from the DBE stage, after RGB Comination and before Photometric Calibration.

Thank you

Peter

[msmythers]

Peter

Take a look at this thread. I think it should help.

https://pixinsight.com/forum/index.php?topic=12366



Mike

[peterbrack]

Thanks Mike

One says try ABE, which I will try, instead of DBE.

Some are about multiple passes of DBE, as well as looking at sample weights.

I will try these and post back

Peter

[John_Gill]

Hi,

Perhaps try "Division" instead of "Subtraction" in DBE and ABE.  I have done both on an image with good results.

Look up
John

[peterbrack]

I have tried another image which I am also finding I have the marked colour gradient. I have used the setting as above, increasing threshold in DBE to 1.3. Have attached a screenshot of DBE settings

Here is a dropbox link to the post DBE/ Photometric Calibration

https://www.dropbox.com/s/93l7jnx1yj8m53h/Image05_DBE_PhotCal.xisf?dl=0

Here is a link to the RGB combination ( so without DBE/PC)

https://www.dropbox.com/s/uudr9mpqax5845u/RGBComb.xisf?dl=0

Would appreciate any guidance in how to develop it better.

Thanks

Peter

[RickS]

Check the edges of the image carefully and look for areas that need to be cropped off before running DBE.  If you automatically generate samples they will start and finish at the outside edges which is not a good thing if you have crappy edges due to deliberate or accidental dithering.

Cheers,
Rick.

[peterbrack]

Thanks Rick for your reply.

I did do a crop on all sides, esp at the top which needed more. Maybe I should move the points away from the edge that are there?

Here is  a link after HT to take nonlinear

https://www.dropbox.com/s/fec2nc1dkkj5mi4/Image05_DBE_Hist.jpg?dl=0

[RickS]

Hi Peter,

I noticed that the left hand edge had an area that was quite ragged too.  I didn't play with the data much as my Internet link is currently saturated and an upload will just slow things further...

Cheers,
Rick.

[peterbrack]

You're right Rick, yes I didn't see that. You have better eyesight than me. I am not doing myself any favours as editing on a Mac laptop at present and so image size is an issue. But can see the band down the left side . Will try removing that and give another go.

Peter

[peterbrack]

I have just done a run taking out the left hand side and a little of each other side. I also went through every point and moved any that had Wr Wg Wb readings lower than 0.75

I applied the same DBE for the RGB to the Luminance, then HT both then LRGB Combination.

Of the screenshot I posted, the left image is what I had before after LRGBCom. The right image is after the further dynamic crop and DBE changes.

Peter

[peterbrack]

This is the same two images in the last post, but taken through further development. The one on the left, which I didn't crop further and didn't fix up the DBE has less red gradient in it than the one I did make improvements on, which is puzzling to me.

Peter

[RickS]

Hi Peter,

Attached is a zip file with a couple of process icons. Use the DynamicCrop first and then apply the DBE two or three times (don't drag & drop - select the target image and then double click the DBE icon.)  It seems to do a reasonable but not perfect job as far as I can tell on my crappy laptop screen.  I'd consider it good enough.  With a little background noise reduction and a gentle stretch it should look fine but you can always mask protect the foreground and use CurvesTransformation to drop the saturation too.

Cheers,
Rick.

[peterbrack]

Hi Rick

Thank you so much for taking the time to do that, really appreciate it and was interesting to see the settings you made. Am unsure how you decided on the points, would be interested to know.

I made the changes and the DBE image looked a lot better. I then did PhotometricColourCalibration and looked okay. After MLT for noise reduction I went to make non linear- this where I had an issue. I have always done an automatic stretch with the ScreenTransferFunction, and then applied that to HT and then HT to the image. After PCC, I cancelled the STF stretch, and then repeated it, something I do , unsure why, anyway it showed up a lot of colour gradients which was disappointing. So I was thinking maybe this method of doing things is not the best way.

I then decided to try MaskStretch instead, and that produced a much better result as far as gradients. Then image wasn't quite as bright, but looked better I thought. I'm now questioning to the best way to make non-linear, as I think it has been causing me a lot of problems.

Anyway, I then used TGVdenoise, HDRMT, LH, MT, and curves to produce the image in the attached screenshot, which is a lot better than how things turned out previously.

Any comments would be appreciated.

Thanks again Rick

Peter

[RickS]

Hi Peter,

I try to go for a minimal set of sample points because a complex gradient probably includes some of your data.  I'll often start with just 4 samples in the corners and one at the midpoint of each edge.  In this case I added a point in the middle of each quadrant and then, because I was still getting some red on the right hand side I placed a few extra points to catch either side of where it wax appearing which cleaned it up a bit better.

The default STF stretch can be too strong.  I always stretch manually.  Unless you have a *lot* of data the background will usually have warts that are visible if you stretch it too far.  It may help to use a "S" curve with RGB/K in CurvesTransformation to darken the background. Applying a mask and desaturating the background is sometimes helpful too.

Where are you imaging M83 from?  I'm working on the same target right now...

Cheers,
Rick.

[peterbrack]

Hi Rick

I am imaging from the south west of Australia. Yourself? What exposure length are you using. I was using 600s. Have only just started autoguiding.

Do you use HistogramTransformation for your stretch  and that manually? Yes I agree the default STF can be too much, I think it has caused problems with a lot of my images.

Do you ever use mask stretch? Haven't a lot of experience  with it.

Yes I did put an S curve with CurvesTransformation

Will try out your sample point method too. I notice you put use a large sample radius. Does that help smooth out  the gradient.

Peter