Stretch before Processing

[John_Gill]

Hi,

I am using Canon DSLR camera.  I setup all my images in BatchPreProcessing and it completes perfectly using all the default settings. PixInsight produces a "light-BINNING_1" file.  I then Crop and run DynamicBackgroundExtraction and this produces a decent image.  I then run "Atrous" or "MultiscaleLinearTransformation" or "MultiscaleMedianTransformation" or any other function.  Now my image is starting to take shape and everything is still linear.  I now run HistogramTransformation & ScreenTransferFunction with the defaults and the image now turns to a horrible mottled color. 

My only way around this is to "Stretch" the image before applying the noise reduction and other functions.  This means that I have to convert my image to non-linear before doing any processing..... Something is not right and it is most likely me...

Any help in this regard will be much appreciated.
Regards
John

[ChoJin]

It would help if you could post some images to illustrate the issue

Also you have to be gentle with noise reduction at the linear stage.

[John_Gill]

Hi,

Attached is the resultant image after a very rough DBE and MLT and then HT & STF.

Regards
John

[jkmorse]

John,

Anything you do to process images in the linear stage requires a very delicate hand and, unless you have great SNR, can do exactly what you are seeing.  The problem is that you end up creating a mess in the linear stage that gets exaggerated when you do your stretch.  In the first instance, you should never run Atrous or MMT in the linear phase and, depending on your data, even go easy with MLT.  Given good SNR you can also apply a little TGVD in the linear phase, but even with good data that can be tricky.  Given what you are seeing, I would strongly recommend doing only minimum processing in the linear stage beyond maybe DBE and background neutralization and color correction.  Save everything else until after you stretch the data.  You can then clean up your noise with TGVD.  The key is that what looks like very minor variations in the linear phase get stretched to a mottled mess when you go non-linear.

If possible, try sending a link to the image that comes out of BPP, but before you do anything else so we can see and play with the data.

For what it's worth,

Best,

Jim

[vicent_peris]

Hi John,

The processes that should be always applied to linear images are these ones:

- Preprocessing.
- Gradient correction.
- Background neutralization.
- Color calibration.


These are processes that cannot be avoided and should be applied always before stretching. If you want to denoise the image, please follow this advice:

- Denoising a linear image can be very powerful, but requires a lot of experience and very delicate work. I would start by first stretching and the applying a denoising process; think of it as a way to train you eye, since denoising is very difficult on linear images because the details are still not enhanced.

- Denoising a non-linear image usually requires an inverted lightness mask to protect the highlights, since the noise is stronger on the shadows. If you don't apply any mask, you'll be surely forced to erase valuable details in the highlight in order to decrease the noise level in the background. You can simply output the lightness of the color image with ChannelExtraction and load it as a mask on the color image.

- A good way to start is always TGVDenoise. Play with the Edge protection parameter to adjust the denoising effect.

- If you want to completely remove the noise, you'll surely fail. It's only my opinion, but my approach is to reduce the noise, not to erase it.


HTH.
Best regards,
Vicent.

[John_Gill]

Hi Jim and Vincent,

Thank you for the wonderful advice on processing.  I will definitely redo this image and see what details I can get out. 

Look up
John