Author Topic: Telea inpainting  (Read 5148 times)

Offline mmnb

  • Member
  • *
  • Posts: 74
Telea inpainting
« on: 2019 December 05 21:27:04 »
Coming back from  J-P Metsavainio's talk at AIC, I really liked the idea of separating stars and working on nebulosity.  Adam Block was also talking about his star shrinking method which involves removing stars and then MMT to "inpaint".

I was just playing with OpenCV's inpainting using PI to create the star subtracted image and star mask (created with MLT and some dilation):
 https://www.learnopencv.com/image-inpainting-with-opencv-c-python/


It seems that opencv's Telea method does a pretty good job (takes a ridiculously long time though, several hours on my pokey 2015 mac airbook for a 2x drizzled image (originally 10k x 7k pixels) .  Nothing great about the m31 image (just the last thing I took/processed), but the star traces seem a touch better with open cv...I think the larger stars need masks that are a bit bigger (they leave bigger traces in the bg).   I'm no expert here, just wondering if anyone might be interested in having this in PI.  Masking the bigger stars with bigger masks for inpainting would improve thing a bit more I think.

(big pngs)
Original:
https://drive.google.com/open?id=1T1TAUwjb_4iB3KXBocnygmt74J9ZNqTy

Starnet (default params):
https://drive.google.com/open?id=13mkj5TGKb7PGvz_UyuoCi51ocuKVcGkH


OpenCV inpaint:
https://drive.google.com/open?id=12uEw5qmw0WQmftcdLtd9MXTd2Tugprmx

EDIT: zooming in, Telea sure does leave very "radial' patterns....from far away, the filled in patterns look lower contrast to me though...
« Last Edit: 2019 December 05 22:23:56 by mmnb »