›

Gradient removal. Best on non linear?

Benoit Saintot
in General
Hello

I have a friend, who is quite experienced and produces nice images, who tells me that gradient removal should be done on non linear images, especially for narrowband, because at the linear stage, the background resides only in very few levels, and it might result in "steps" ( maybe posterizarion is the correct word, I'm French) during stretching.
I'm quite surprised, because I always read that this step should be done at first stages of processing, during linear phase.
I gave a try to a couple of processes, like Graxpert. DBE/ ABE, GradietXtemrinator, some work on linear, some on both...

So what do you think about this?

Thanks a lot.

Comments

  • BlockHeadBlockHead
    Numbers are numbers. They can indeed be small and not DISPLAY correctly (quantized)... the VALUES in the linear data (32-bit) are not quantized however. 
    In addition, good integrated images (enough collected photons) do have sky values (backgrounds). As long as you have proper calibration (with a pedestal if necessary)- all information is retained and there should not be a loss of any information. 

    Applying DBE at the linear stage has many benefits. One is that you can do this and not interfere with color calibration (or color mapping in the case of NB). If you do DBE on a non-linear image- you will affect the colors you are working on strongly. DBE on non-linear images is a VERY LARGE effect because the values (histogram) fills more of the tonal space. DBE on linear images really will only effect the background- because these values are small and you choose them. 

    I am tired of typing... so there are some other reasons as well. 
    But more importantly did you observe quantized VALUES in the data? 

    Remember when you use DBE you can always normalize the output. This will prevent subtraction (for example) from actually giving you too small values. 

    Be certain to write back with specific examples of where my information above is not applicable.

    -the Blockhead
  • Benoit Saintot
    Hi Adam

    Thanks a lot for your reply.

    Personnaly, I never noticed quantized data's or such an effect, hence my question.
    I also only shoot mono, so I'm not really concerned by the color gradients.

    I will think about it, but as long as it works for me, I think I will stick with the linear gradient removal.
    By the way, Inm really looking forward Russ's Gradient Removal AI. DBE is a bit old I find, it suffers comparison now with GraXpert or tools that use RBF ( Siril/ GarX) / Kringin or AI functions ( Graxpert)
    For GradientXtelrinator, I don't know the maths behind.

    Thanks Adam.

    Benoit.

Sign In or Register to comment.