Getting Deep on Processing Math

Since we’ve been going back and forth to the LRE house and I haven’t had a chance to get things fully set-up in LA for CCD imaging, my main target has been the Moon. As in the previous post, I have tried additional mosaics. I have also kept experimenting with processing Lunar images.

My basic process in PixInsight is:

  1. Convert the image to 32 bit
  2. Upsample by 200%. Deconvolution simply won’t work on the original pixel depth.
  3. Use regularized Van Cittert algorithm deconvolution to sharpen the image
  4. Flatten the image with HDR Wavelet transform at the default settings
  5. Adjust the contrast of the image with curves
  6. In some cases, do some minor noise reduction with GreyCStoration.

Most of my time is taken tweaking the deconvolution parameters. (If you are starting on this, the PixInsight team has a great processing example on their site.) This time, I learned that other parameters matter too.

In iterating on the deconvolution, I noticed that this image had more noise that other recent ones. The deconvolution was bringing it out clearly. After several hours, I figured out what I had done differently. On my prior Lunar images, I had selected bicubic b-spline interpolation rather than default when I upsampled the image. No particular reason, but I picked that instead of the default. This time, I had used the default (“automatic”) which I believe chose bilinear interpolation.

No, I don’t understand the difference in the algorithms, but I know they are different. Time do do some reading on the subject.

So I experimented and learned that for my Lunar images, bicubic b-spline is sharp, cubic b-spline produces less noise, and the other offered choices aren’t worth using. And I learned that you can go past critical details in processing your image without even knowing it. Layers. (Onion layers, not parfait layers, if you follow the reference.)