Extract specific edges (e.g. only those of circles or straight lines)

[piw-bot]

First check

• I added a very descriptive title here.
• I used the GitHub search to find a similar question and didn't find it.
• I searched the PyVista documentation, with the integrated search.
• I already searched in Google "How to X in PyVista" and didn't find any information.
• I already read and followed all the tutorial in the docs and didn't find an answer.
• I will NOT make duplicate discussions or issues in the future.

Commit to Help

• I commit to help with one of those options.

Sample Code What is the problem, question, or error?

Write a short description telling me what you are doing, what you expect to happen, and what is currently happening.

def edge_detection():
    mesh = pv.read("my_mesh")
    edges = mesh.extract_feature_edges(45)

    p = pv.Plotter()
    p.add_mesh(mesh, color=True)
    p.add_mesh(edges, color="red", line_width=5)
    p.show()

    #TODO Extract only specific edges, e.g. only the edges of a circle

Description

Hi, I extracted the edges of the mesh. How can I extract only specific kind of edges, for instance only those of the circle? How can I define which edges to extract? Or is there a possibility to extract all the edges and then filter them afterwards?

System Information

--------------------------------------------------------------------------------
  Date: Wed May 08 15:04:12 2024 Mitteleuropäische Sommerzeit

                OS : Windows
            CPU(s) : 19
           Machine : AMD64
      Architecture : 64bit
       Environment : Python
        GPU Vendor : NVIDIA Corporation
      GPU Renderer : Quadro RTX 4000/PCIe/SSE2
       GPU Version : 4.5.0 NVIDIA 535.98
  MathText Support : False

  Python 3.10.0 (tags/v3.10.0:b494f59, Oct  4 2021, 19:00:18) [MSC v.1929 64
  bit (AMD64)]

           pyvista : 0.43.5
               vtk : 9.3.0
             numpy : 1.26.4
        matplotlib : 3.8.4
            scooby : 0.9.2
             pooch : 1.8.1
            pillow : 10.3.0
           IPython : 8.23.0
        ipywidgets : 8.1.2
             scipy : 1.13.0
      nest_asyncio : 1.6.0
--------------------------------------------------------------------------------

Screenshots

[darikg]

Assuming those circles are disconnect from other edges, after extracting the edges, try applying the connectivity filter to label disjoint boundary loops and extract them by their RegionId

[piw-bot]

Thank you for your help, unfortunately everything is connected as you can see in this screenshot

[user27182]

This is similar to an edge detection / feature extraction problem. This is common in computer vision tasks and is not trivial to do. If the input was image data, you could use something like the Hough transform to extract lines and circles (this is just an example, there are no vtk/pyvista methods that implement the Hough transform as far as I know, and your input is not image data anyway).

For this mesh, what features of the circles and lines make them distinct from other parts of the mesh? Well, the circle bits are curved and the flat bits are not. So the curvature filter may help extract the curved (circle) features.

Have a look at the bottom of the connectivy examples linked by @darikg above. There is an example on how to extract regions from curvature using seed points.

E.g. if we have this mesh and extract its edges (circles and lines):

from pyvista import examples
import pyvista as pv
import numpy as np
mesh = examples.download_cad_model()
feature_edges = mesh.extract_feature_edges(80, boundary_edges=False, non_manifold_edges=False, manifold_edges=False,)
feature_edges.clear_data()
feature_edges.plot(render_lines_as_tubes=True, line_width=10)

We can plot its curvature, e.g.:

mesh['curvature'] = mesh.curvature('minimum')
mesh.plot(
    clim=[-4, 4],
    cmap='bwr',
    below_color='blue',
    above_color='red',
)

This will output the image below. You can see that the circles we want to extract are in blue, and have distinct values from non-circles. You can experiment with different curvature measures (e.g. maximum, minimum, etc.) and different ranges, e.g. clim=[-10, 10], clim=[-1, 1], etc.

You can then use the connectivity filter to extract only the blue regions using a scalar range, and then use extract_all_edges.
The idea here is we want to extract only the "really" curved edges, the ones that are definitely part of the circles we want to extract. Here the range [-np.inf, -4] is used to define "really curved", but again you can experiment with different values.

curved_edges = mesh.connectivity(scalar_range=[-np.inf, -4]).extract_feature_edges(80, boundary_edges=False, non_manifold_edges=False, manifold_edges=False,)
curved_edges.plot()

These really curved edges can then be used to "seed" the connectivity filter, i.e. use the points from our "really curved" lines as seed to find all the other curved lines connected to the really curved ones.

First we split each of our "really curve" lines from above into separate meshes using split_bodies.

seed_meshes = curved_edges.split_bodies()

Then, use a seed point from each curved line and get all other, slightly less curved parts attached to the seed meshes.
Whereas before we used [-np.inf, 4] to get the really curved parts, here we relax that to [-np.inf, -2] to still get curved parts, but allow for slightly less curved sections. Again, you will want to experiment with different values.

circle_meshes = [mesh.connectivity('closest',seed_mesh.points[0], scalar_range = [-np.inf, -2]) for seed_mesh in seed_meshes]

The above generates a list of meshes, each mesh containing (hopefully) one circle each. We merge them into a single mesh using a MultiBlock to combine them.

merged_circles = pv.MultiBlock(circle_meshes).combine()

The extracted meshes will contain both cells and points, i.e. the curved regions. Use extract feature edges again to only show the circles.

circles = merged_circles.extract_feature_edges(70, boundary_edges=False, non_manifold_edges=False, manifold_edges=False,)
circles.plot()

You can see that the circles are extracted, though the result is not perfect and there are some straight lines present. You can try using the connectivity again to remove the noise, (again, see connectivity examples)

Another idea is to use features like gradient instead of curvature, and try to follow a similar process used by Canny edge detection.

[darikg]

in addition to @user27182's excellent suggestions I would also suggest trying extract feature edges. Since the circular parts meet the flat parts at 90 degree angles, that should be able to isolate the circles as disconnected loops.