- #1
- 7,713
- 3,835
There seems to be enough of us night owls who use DSS for me to open a discussion about how to 'tune' all of the parameters to get an optimal result. Personally, I've used the program for a few years but still have no clue about many of the settings, and would appreciate any insight. First, what I've learned:
In general, stack RAW rather than JPG: the extra bits in RAW help to preserve the dynamic range and color saturation. Using 'flats' helps enormously, but remember to acquire the flats at the same f/# as the images, otherwise the calibration will spectacularly fail because the vignetting can change dramatically. I can stack images taken with high ISO, as long as there's enough of them to smooth out the noise. I do not use any in-camera noise reduction, and set the white balance to 'auto'.
When registering the images, I get optimal results when my threshold provides about 50-100 stars per image. More stars just takes longer to align, fewer stars results in dropped frames. After I register the images, I manually threshold the list by star FWHM (full width half maximum)- frames above the threshold are smeared and/or blurred, so I just delete those from the disk. The specific threshold varies, but for me it's around 5.5 pixels.
For stacking parameters, I've gotten the best results with 'auto adaptive weighted average', with a number of iterations about equal to the number of images. I have tried the 2x and 3x dithering with marginal improvements- I have to window the images to do this, and batch stacking does not seem to remember that, leading to failed (out of memory) runs.
After I adjust the initial 32-bit image, I export the file and perform additional post-processing with ImageJ and Neat Image as needed (e.g. background subtraction, noise reduction, color balance, etc.)
Now what I would like to learn:
1) If my field of view has a large dynamic range (bright stars through faint nebula), how can I best preserve the dynamic range? I've tried acquiring a subset of images at low sensitivity to not overexpose the highlights, but then there's not enough stars for DSS to align. Alternatively, after stacking, I can adjust the exposure curve of the 32-bit image to almost look like a step function (more on this later), but then my bright stars bloom out to giant blobs. Any tips?
2) After stacking, I have a 32-bit image that I first manipulate in DSS and save as a 3 x 16-bit TIFF stack. For the initial DSS post-processing step, I have no idea how to optimally white balance and luminance adjust- for example:
The RGB/K levels can be scaled linearly, logarithmically, log(log),... Which is the best? Is there a best?
My initial color balance method is to adjust the color sliders until the gaussian curves are all centered on each other, with the peak located at about the 25% level. This is not easy, I end up moving the mouse in millimeter increments to adjust, so I'm wondering what I'm doing wrong.
When I adjust the luminance, I often end up with a *very* steep midtone to pull out the faint stuff, also requiring very fine mouse movements- but this is because I am viewing the image on an 3 x 8-bit RGB display, which has much less dynamic range than the data. Hints?
This post is getting long... that's enough for now.
In general, stack RAW rather than JPG: the extra bits in RAW help to preserve the dynamic range and color saturation. Using 'flats' helps enormously, but remember to acquire the flats at the same f/# as the images, otherwise the calibration will spectacularly fail because the vignetting can change dramatically. I can stack images taken with high ISO, as long as there's enough of them to smooth out the noise. I do not use any in-camera noise reduction, and set the white balance to 'auto'.
When registering the images, I get optimal results when my threshold provides about 50-100 stars per image. More stars just takes longer to align, fewer stars results in dropped frames. After I register the images, I manually threshold the list by star FWHM (full width half maximum)- frames above the threshold are smeared and/or blurred, so I just delete those from the disk. The specific threshold varies, but for me it's around 5.5 pixels.
For stacking parameters, I've gotten the best results with 'auto adaptive weighted average', with a number of iterations about equal to the number of images. I have tried the 2x and 3x dithering with marginal improvements- I have to window the images to do this, and batch stacking does not seem to remember that, leading to failed (out of memory) runs.
After I adjust the initial 32-bit image, I export the file and perform additional post-processing with ImageJ and Neat Image as needed (e.g. background subtraction, noise reduction, color balance, etc.)
Now what I would like to learn:
1) If my field of view has a large dynamic range (bright stars through faint nebula), how can I best preserve the dynamic range? I've tried acquiring a subset of images at low sensitivity to not overexpose the highlights, but then there's not enough stars for DSS to align. Alternatively, after stacking, I can adjust the exposure curve of the 32-bit image to almost look like a step function (more on this later), but then my bright stars bloom out to giant blobs. Any tips?
2) After stacking, I have a 32-bit image that I first manipulate in DSS and save as a 3 x 16-bit TIFF stack. For the initial DSS post-processing step, I have no idea how to optimally white balance and luminance adjust- for example:
The RGB/K levels can be scaled linearly, logarithmically, log(log),... Which is the best? Is there a best?
My initial color balance method is to adjust the color sliders until the gaussian curves are all centered on each other, with the peak located at about the 25% level. This is not easy, I end up moving the mouse in millimeter increments to adjust, so I'm wondering what I'm doing wrong.
When I adjust the luminance, I often end up with a *very* steep midtone to pull out the faint stuff, also requiring very fine mouse movements- but this is because I am viewing the image on an 3 x 8-bit RGB display, which has much less dynamic range than the data. Hints?
This post is getting long... that's enough for now.
Dave