Hello dear RG community.
I'm trying to identify contours of bubbles on the images taken with Nikon 1 J4 camera ( backlit through a diffuser with a halogen lamp). I use Python.
I'm trying to go with Watershed, but both opencv and skimage make use of binary markers to assist Watershed.
Binarization of bubble images doesn't suit that purpose well. Moreover, if I was able to obtain good results I would just go with binarization instead of finding contours.
I haven't been able to find Watershed codes without markers.
I'm wondering if anybody has an open source code for Watershed without markers.
And in general, I'm wondering if anybody has come up with a code to identify bubbles contours or just bubbles in a wide range of two-phase flow regimes and if the code is open source.
Thank you in advance.
Ivan
As far as I understand, markers is a necessary element in Watershed. See Wikipedia: https://en.wikipedia.org/wiki/Watershed_(image_processing).
As stated in Wikipedia, obtaining markers can be achieve either by gradients or by binarization.
For binarization, here are a couple of posts: https://pyimagesearch.com/2015/11/02/watershed-opencv/ and https://opencv24-python-tutorials.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_watershed/py_watershed.html
For gradients, here's a post: https://scikit-image.org/docs/dev/auto_examples/segmentation/plot_watershed.html
To my experience, gradients based markers give better results for a wide variety of two-phase flow regimes (from dispersed bubbly flows to churn flows), but it is difficult to get the parameters right.
I have halted my work on image segmentation in a wide variety of two-phase flow regimes for now. I'm concentrating on dispersed bubbly flow regimes now, where simple binarization with subsequent contours finding works pretty much fine.
Still, I'll appreciate if anybody has ideas on Watershed.
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators