Persistent-based segmentation

Dmitriy Morozov

I think you'll have to explain better what you are trying to do before I can help you.

You can get cycles (which are not unique) from Dionysus. That question has come up before on this list. For example, see

I'm not sure what you mean by pixels having persistence larger than a given threshold.


On Wed, Jun 5, 2019 at 9:57 AM Giacomo Nardi <giacomo@...> wrote:


I am using the dionysus package to compute persistent homology for images.  I was wondering if there is any function to perform segmentation based on persistence.

How can I select cycles/pixels having persistence larger than a given threshold ?

Thank you

Giacomo Nardi | Research Scientist TDI
Mauna Kea Technologies | 9 rue d Enghien – 75010 Paris 
Tel : + 33 1 48 24 03 45 | Fax : + 33 1 48 24 12 18 |
Cellvizio iPhone App | Follow us on Twitter | Follow us on LinkedIn 


This message and any attached documents are CONFIDENTIAL and intended only for the use of the intended recipient.
Ce message et tous les documents attachés sont CONFIDENTIELS et à la seule attention du destinataire pour lequel ils ont été écrits.

Jack <giacomo@...>

Hi Dmitriy,

did you have any time to think about my question?


Dmitriy Morozov

I don't really understand your question. The linked messages give a way to get actual cycles (for example, sequences of edges, not just births/deaths). Do you want to do something else?

But I'm really not sure what you are asking. What is a "border of a cycle"? What are the pixels of the images belonging to a cycle? (Other than getting the cycles as described in the messages I linked.) It's also worth noting that there is no uniquely defined cycle that persistence finds. I think you need to try to formalize mathematically what you want to find, then it would be possible to figure out if it's something Dionysus computes.

I also don't understand the strategy you describe: if you want all pixels between value 5 and 22, you don't need Dionysus for that; NumPy can give you these pretty quickly.

In short, I don't think there is an optimized function in Dionysus to do what you want, besides extracting the representative cycles, as described in the messages I linked originally.


On Thu, Jun 13, 2019 at 5:20 AM Jack <giacomo@...> wrote:
Hi Dmitriy,

did you have any time to think about my question?