Ok ok, here I am.
As a personal philosophy, I am strongly against rigid workflows, becouse image processing is not a cooking recipe. It is far more important knowing what the different tools (processes) do, and why we should use them, so we call them if we feel that it is the right time. A image is like a sculpture, we follow a path looking at the material, and bringing the artwork to the surface :)
Well, having stablished that, I divide my workflow into several groups:
- Calibration
The main idea behind this is to prepare data to the non linear modifications (histogram stretch). So, the word calibration is taking in a broader acception than just bia-darks-flats.
All the processes applied at this stage are done to the linear data.
Let's menction a few steps that we may follow at this stage:
If we have calibration images (bias, darks and flats), apply them (I'll asume that you know how to do that with PixelMath, or use DeepSkyStacker, for example).
Align and stack all the frames.
Crop the result to the intersection of the frames (to avoid strange pixels at the boundaries).
Use the CloneTool to erase hot pixels, cosmic rays, etc.
Deconvolucionate data. This is considered as a calibration process, becouse the idea behind deconvolution is to restore data that was "blured" becouse of atmospheric turbulence or lens distortions. Iw we deconvolutionate color data, a different PSF is likely to be found, so I think that it is best to apply it to each filter data in separate images.
Color calibration and LRGB combine.
Deal with gradients (either DBE or ABE). Depending on the source, divide or substract the model.
- Main Adjustments
The next stage contains all the color, contrast and brightness adjustments that will be applied to the image, turning it close to the final look.
We start applying the HistogramTransform, or the AutoHistogram processes, to define the dynamic range limits. This means, set the black and white point. As a general recomendation, I suggest not to clipp a considerable amount of data. Since noise is stronger at the shadows, I ussually clip up to a 0.015% there. Hihghlights, by the other hand, are almost pure data, so no clipping or a very low one should be used. My "magical number" is 0.005%. It works well with film data :) Then, with either of those processes, raise the middtones balance so the background turns into the "correct" brightness range. You may neutralize the background at this step.
Next, comes all the curves adjustments. Slight R, G or B changes should be applied if a color balance is needed. The brightness and contrast adjustment is performed using the L channel. Color saturation is better handled with the c channel.
After those two fundamental processes, starts the fun ;) There are many ways to further modify the image (ExponentialTransforms and ColorSaturation, for example). Make use of masks if needed.
- Noise Reduction
Simple put, make use of PI's arsenal to improve SNR ratio. Don't be too aggressive, but try to preserv all the structures. We may use different tools, and iterate, to get the desired result. Don't try to speed up things and be done with a unique application/process.
Usually my workflow at this stage is:
GRAYCStoration, with a inverted luminance mask.
ACDNR (stronger to the chrominance)
SCNR, to remove any green cast.
A short note about SCNR applied at the end. SCNR is very usefull to remove green hot pixels, and this kind of noise, and also green haloes at stars, due to chromatic aberrations. But, if we apply it before the other techniques, it may change the hue of the background. And, if we have chromatic aberrations, removing the green halo changes the profile of the stars, and theyr color becomes very hard to fix. So, if we left it to the end, data will be smoother, star colors will be more simmetric and ussually the later will not be affected significantly.
After the noise reduction processes, it is likelly that there is a slight contrast loss, and that is a bit of free space at the ends of the dynamic range, so, new histogram adjustments should be applied. Further adjustments are optional.
- Structure Enhancement
Under this I group all the techniques that enhances the structures. High pass filters, new deconvolutions and wavelets.
In my workflow, I separate small scale features from the medium/large ones. I process with wavelets (ATrousWavelets and/or HDRWT) and curves each one os these "subimages". The separation is done with morphological filters.
Again, after combining the different scales, new adjustments may be needed. At this stage of things, they are the final ones, so we should make sure that we are happy with the result.
If needed, a specific star shaping routine may be applied at the very end. This is done with a star mask (we may reuse the small scales subimage we had previously, modifying it with histogram, wavelets and morphological transforms).
So, here are my 10 pesos ;)
