29
$\begingroup$

I am trying to estimate curvatures on a triangulated surface/manifold using the algorithm of : Meyer, M., Desbrun, M., Schröder, P., & Barr, A. H. (2003). Discrete differential-geometry operators for triangulated 2-manifolds. In Visualization and mathematics III (pp. 35-57). Springer Berlin Heidelberg.

This algorithm gives estimates at each vertex on the mesh of the local mean and Gaussian curvature, based on the angles of the surrounding triangles. The algorithm also allows for estimates of principle curvature directions. I am fairly new to Mathematica and am struggling with speeding up the calculations (I have nested For loops which after reading this forum seems to be something that one should avoid in Mathematica). I would greatly appreciate any help in where I could speed up the code. I hope also that this code will also be useful to others as well in the long term.

As an example we can take an ellipsoid and discretise it using the BoundaryDiscretizeRegion function:

aellipse = 1;
bellipse = 0.6;
cellipse = 0.3;
a = BoundaryDiscretizeRegion[Ellipsoid[{0, 0, 0}, {aellipse, bellipse, cellipse}],
                             MaxCellMeasure -> {"Length" -> 0.3}]

Discretized Ellipsoid

Now we scan over every vertex i (and then every vertex j neighbouring vertex i) giving an estimate of mean and Gaussian curvature:

ℛ = a;
(*Angles at each triangle*)
va = VectorAngle[#1 - #2, #3 - #2] & @@@ Partition[#, 3, 1, {2, -2}] & /@MeshPrimitives[ℛ, {2}][[All, 1]];
(*Coordinates of Mesh*)
mc = MeshCoordinates[ℛ];
(*Number of vertices*)
nvert = MeshCellCount[ℛ, 0]; 
(*Number of faces*)
nfaces = MeshCellCount[ℛ, 2];
(*Using props values of mesh to calculate list of areas (sum should be the same as our voronoi list minus boundary areas)*)
Areasoftriangles = PropertyValue[{ℛ, 2}, MeshCellMeasure]; 
(*Number of nearest neighbours data table over all vertices*)
nnbrs = Table[{}, {ii, 1, nvert}];
(*Mean Curv data table over all vertices*)
H = Table[{}, {ii, 1, nvert}]; 
(*Gaussian Curv data table over all vertices*)
K = Table[{}, {ii, 1, nvert}];
(*Normal Vector data table*)
Nvectors = Table[{}, {ii, 1, nvert}];
(*Area around vertex data table over all vertices*)
Acalc = Table[{}, {ii, 1, nvert}];
(*List of labels showing we are on a boundary, to be set to 1 if we are on the boundary*)
blist = Table[{}, {ii, 1, nvert}];
(*List of labels of triangles and positions in the list at which the vertices are obtuse*)
obtusetrianglelist = Position[va, n_ /; n > π/2]; 
(*List of labels of only the triangles at which the vertices are obtuse*)
obtusetrianglelisttrinum = obtusetrianglelist[[All, 1]];

For[i = 1, i < nvert + 1, i++,
(*Starting vector for mean curvature sum*)
MeanCVect = {0, 0, 0};
(*Counting value for Voronoi*)
AMixed = 0;
(*Sum of errors to calc principle directions*)
Esum = 0;
(*Test value to see whether a given vertex is on the edge of surface \
(i.e. we are evaluating an open surface)*)
edgetest = 0;
(*List of edges attached to Point i*)

bb = Select[MeshCells[ℛ, 1][[All, 1]], #[[1]] == i || #[[2]] == i &];
(*List of other vertices attached to Point i*)

bb1 = Cases[Flatten[bb], Except[i]];
(*Count Number of Nearest vertices on mesh*)
nnbrs[[i]] = Length[bb1];
(*Calculation of Area, Curvature etc at a given vertex*)
For[j = 1, j < Length[bb1] + 1, j++,   
(*Select Point jj in list of other nodes attached to Point i, to be summed over all connected nodes*)
    jj = bb1[[j]];
    (*Select the two triangles that are on either side of this line*)
    cc = Select[MeshCells[ℛ, 2][[All, 1]], (#[[1]] == i || #[[2]] == i || #[[3]] == i) && (#[[1]] == jj || #[[2]] == jj || #[[3]] == jj) &];
    (*Check that there are two triangles, if not we are on a boundary and we will then ignore them in the calculation*)
    If[Length[cc] == 2,
    (* Calculate the position in the list of Triangles where the two triangles attached to the line between i and j are *)    
    d1 = Position[MeshCells[ℛ, 2], cc[[1]]][[1, 1]];
    d2 = Position[MeshCells[ℛ, 2], cc[[2]]][[1, 1]];
    (* Calculate the vertex numbers of the vertices in the triangles opposite to the line ij *)

    ee = Cases[Cases[Flatten[cc], Except[i]], Except[jj]];
    (* Find where this is in the list of three vertices per triangle*)
    e1 = Position[cc[[1]], ee[[1]]][[1, 1]];
    e2 = Position[cc[[2]], ee[[2]]][[1, 1]];
    (* Calculate the angle based on the vertex number and the triangle number*)
    a1 = Cot[va[[d1]][[e1]]];
    a2 = Cot[va[[d2]][[e2]]];
    (*Calculation of ijvector*)
    ijvect = mc[[i]] - mc[[jj]];
    MeanCVect += (1/2)*(a1 + a2)*(ijvect);
    (*Area calculation, modified Voronoi checking for obtuse triangles*)
    (*In this first version we will double our calcs, as triangles will be tested twice whether they are obtuse or not*)

    If[MemberQ[obtusetrianglelisttrinum, d1],
    (*Now do test to see which triangle area we add*)     
        ObtVnum = Position[obtusetrianglelisttrinum, d1][[1, 1]];
        Vnum = cc[[1, obtusetrianglelist[[ObtVnum, 2]]]];
        If[Vnum == i,
            (*Triangle Obtuse at i, therefore add half of area T/2*)

            AMixed += (1/2)*(1/2)*Areasoftriangles[[d1]];
            ,
            (*Triangle Obtuse but not at i, therefore add half of area T/4*)

            AMixed += (1/2)*(1/4)*Areasoftriangles[[d1]];
    ]
    ,
    AMixed += (1/8)*(a1)*(Norm[ijvect])^2
    (*If False we add the normal voronoi*)
    ];
    (*Repeat the test for the other angle*)

    If[MemberQ[obtusetrianglelisttrinum, d2],
    (*Now do test to see which triangle area we add*)

        ObtVnum = Position[obtusetrianglelisttrinum, d2][[1, 1]];
        Vnum = cc[[2, obtusetrianglelist[[ObtVnum, 2]]]];
        If[Vnum == i,
            (*Triangle Obtuse at i, therefore add half of area T/2*)

            AMixed += (1/2)*(1/2)*Areasoftriangles[[d2]];
            ,
            (*Triangle Obtuse but not at i, therefore add half of area T/4*)

            AMixed += (1/2)*(1/4)*Areasoftriangles[[d2]];
        ]
    ,
    AMixed += (1/8)*(a2)*(Norm[ijvect])^2
    (*If False we add the normal voronoi*)
 ];

,
(*If the elements are on the boundary we then ignore area and curv calc and set everything to zero*)
edgetest = 1;
blist[[i]] = 1;
Break[];
]
]
If[edgetest == 1,
    (* Set Voronoi Area, mean curvature, and gaussian curvature, to Zero if edge test is 1*)
    AMixed = 0;
    K[[i]] = 0;
    H[[i]] = 0,

(*Calculate Gaussian Curvature*)

    pp = Position[MeshCells[ℛ, 2][[All, 1]], i];
    GaussCAngleSum = (2*π - Total[Extract[va, pp]])/AMixed;
    K[[i]] = GaussCAngleSum;
    H[[i]] = Norm[MeanCVect/AMixed]/2;
    Nvectors[[i]] = (MeanCVect/(AMixed*2*H[[i]]));
    Nvectors[[i]] = Nvectors[[i]]/Norm[Nvectors[[i]]];
];
Acalc[[i]] = AMixed;

];

Now the algorithm seems to work (at least for small meshes ), and gives at least qualitatively results that match analytical values see below:

Comparison of Analytical and Discrete Curvature Values on an ellipsoid

In order to check how robust the algorithm is on other surfaces (and to debug the code) I would like to run it on finer meshes and do it more systematically but as can be expected things slow down very quickly with decreasing mesh size. Any hints would be most welcome.

(I calculated the angles in the triangles based on the post: Angles inside of Voronoi mesh cells

and the measurements of the number of nearest neighbours of each vertex was inspired by : Colouring points in a Delaunay Mesh by the number of nearest neighbours)

$\endgroup$
16
  • 2
    $\begingroup$ Don't use For loops; use vectorized operations. (On a tiny iPod and I'm struggling to type. Perhaps search the site for vectorized" or perhaps you'll get lucky and someone will refactor your code for you.) $\endgroup$
    – Michael E2
    Jun 8, 2016 at 7:12
  • 1
    $\begingroup$ I can't think a better way to slow Mathematica down than forcing it to use indexed procedural methods to compute with arrays. $\endgroup$
    – m_goldberg
    Jun 8, 2016 at 7:43
  • 3
    $\begingroup$ @m_goldberg I don't have an efficient code for curvature computation, and I would definitely benefit from having such a code; I am willing to put a bounty on it if it remains open. $\endgroup$
    – user21
    Jun 9, 2016 at 11:53
  • 1
    $\begingroup$ @m_goldberg, I know for a fact that this sort of computation is often done for meshes; I am in fact working on a solution, and will eventually post something once I'm done. $\endgroup$ Jun 9, 2016 at 12:25
  • 1
    $\begingroup$ Late to the party, but adding in that my lab could definitely use code like this as well. We have something, but it would be nice to have something in mathematica. $\endgroup$ Jun 13, 2016 at 19:55

4 Answers 4

32
+250
$\begingroup$

Note added 1/29/2020: the routines here have a bug where the mean curvature is sometimes computed with the opposite sign. I still need to work on how to fix this.


I guess I should not have been surprised that there are actually many ways to estimate the Gaussian and mean curvature of a triangular mesh. I shall present here a slightly compacted implementation of the MDSB method, with a few of my own wrinkles added in. I should say outright that the current implementation has two weaknesses: the determination of the ring neighbors of a vertex is slow, and that it is only restricted to closed meshes (so it will not work for, say, Monge patches, which are generated by $z=f(x,y)$). (I have yet to figure out how to reliably detect edges of a triangular mesh, since the underlying formulae have to be modified in that case.)

Having sufficiently (I hope) warned you of my implementation's weak spots, here are a few auxiliary routines.

As I noted in this answer, the current implementation of VectorAngle[] isn't very robust, so I used Kahan's method to compute the vector angle, as well as its cotangent (which can be computed without having to use trigonometric functions):

vecang[v1_?VectorQ, v2_?VectorQ] := Module[{n1 = Normalize[v1], n2 = Normalize[v2]}, 
       2 ArcTan[Norm[n2 + n1], Norm[n1 - n2]]]

ctva[v1_?VectorQ, v2_?VectorQ] := Module[{n1 = Normalize[v1], n2 = Normalize[v2], np, nm},
     np = Norm[n1 + n2]; nm = Norm[n1 - n2]; (np - nm) (1/np + 1/nm)/2]

Let me proceed with an example that is slightly more elaborate than the one in the OP. Here is an algebraic surface with the symmetries of a dodecahedron:

dodeq = z^6 - 5 (x^2 + y^2) z^4 + 5 (x^2 + y^2)^2 z^2 - 2 (x^4 - 10 x^2 y^2 + 5 y^4) x z +
        (x^2 + y^2 + z^2)^3 - (x^2 + y^2 + z^2)^2 + (x^2 + y^2 + z^2) - 1;

dod = BoundaryDiscretizeRegion[ImplicitRegion[dodeq < 0, {x, y, z}],
                               MaxCellMeasure -> {"Length" -> 0.1}]

the example mesh

Extract the vertices and triangles:

pts = MeshCoordinates[dod];
tri = MeshCells[dod, 2] /. Polygon[p_] :> p;

Now, the rate-limiting step: generate a list of all the $1$-ring neighbors for each vertex. (Thanks to Michael for coming up with a slightly faster method for finding the neighbors!)

nbrs = Table[DeleteDuplicates[Flatten[List @@@ First[FindCycle[
             Extract[tri, Drop[SparseArray[Unitize[tri - k],
                                           Automatic, 1]["NonzeroPositions"], None, -1],
                     # /. {k, a_, b_} | {b_, k, a_} | {a_, b_, k} :> (a -> b) &]]]]],
             {k, Length[pts]}];

(I am sure there is a more efficient graph-theoretic method to generate these neighbor indices (using e.g. NeighborhoodGraph[]), but I haven't found it yet.)

After that slow step, everything else is computed relatively quickly. Here is how to generate the "mixed area" for each vertex:

mixar = Table[Total[Block[{tri = pts[[Prepend[#, k]]], dpl}, 
                          dpl = Apply[(#1 - #2).(#3 - #2) &,
                                      Partition[tri, 3, 1, 2], {1}]; 
                          If[VectorQ[dpl, NonNegative],
                             ((#.# &[tri[[1]] - tri[[2]]])
                              ctva[tri[[1]] - tri[[3]], tri[[2]] - tri[[3]]] + 
                              (#.# &[tri[[1]] - tri[[3]]])
                              ctva[tri[[1]] - tri[[2]], tri[[3]] - tri[[2]]])/8, 
                             Norm[Cross[tri[[2]] - tri[[1]], tri[[3]] - tri[[1]]]]/
                             If[dpl[[1]] < 0, 4, 8]]] & /@ Partition[nbrs[[k]], 2, 1, 1], 
                    Method -> "CompensatedSummation"], {k, Length[pts]}];

(Thanks to Dunlop for finding a subtle bug in the mixed area computation.)

The Gaussian curvature can then be estimated like so:

gc = (2 π - Table[Total[Apply[vecang[#2 - #1, #3 - #1] &, pts[[Prepend[#, k]]]] & /@
                        Partition[nbrs[[k]], 2, 1, 1], Method -> "CompensatedSummation"],
                  {k, Length[pts]}])/mixar;

Computing the mean curvature is a slightly trickier proposition. Even though the mean curvature of a surface can be positive or negative, the MDSB method only provides for computing the absolute value of the mean curvature, since it is generated as the magnitude of a certain vector. To be able to generate the signed version, I elected to use a second estimate of the vertex normals, and compare that with the normal estimate generated by the MDSB method to get the correct sign. Since the vertex normals will be needed anyway for a smooth rendering later, this is an acceptable additional cost. I settled on using Max's method:

nrms = Table[Normalize[Total[With[{c = pts[[k]], vl = pts[[#]]}, 
                                  Cross[vl[[1]] - c, vl[[2]] - c]/
                                  ((#.# &[vl[[1]] - c]) (#.# &[vl[[2]] - c]))] & /@
                             Partition[nbrs[[k]], 2, 1, 1],
                             Method -> "CompensatedSummation"]], {k, Length[pts]}];

Finally, here is how to compute the estimated mean curvature:

mcnrm = Table[Total[Block[{fan = pts[[Prepend[#, k]]]},
                          (ctva[fan[[1]] - fan[[2]], fan[[3]] - fan[[2]]] + 
                           ctva[fan[[1]] - fan[[4]], fan[[3]] - fan[[4]]])
                          (fan[[1]] - fan[[3]])/2] & /@ Partition[nbrs[[k]], 3, 1, 2], 
                    Method -> "CompensatedSummation"], {k, Length[pts]}]/mixar;

mc = -Sign[MapThread[Dot, {nrms, mcnrm}]] (Norm /@ mcnrm)/2;

To be able to do a visual comparison, I'll derive the analytical formulae of the Gaussian and mean curvature of the surface. The requisite formulae were obtained from here.

gr = D[dodeq, {{x, y, z}}] // Simplify;
he = D[dodeq, {{x, y, z}, 2}] // Simplify;

(* Gaussian curvature *)
gcdod[x_, y_, z_] = Simplify[((gr.LinearSolve[he, gr]) Det[he])/(#.# &[gr])^2];

(* mean curvature *)
mcdod[x_, y_, z_] = Simplify[(gr.he.gr - Tr[he] (#.# &[gr]))/(2 (#.# &[gr])^(3/2))]

Now, compare the results of the estimates and the actual curvatures (coloring scheme adapted from here):

GraphicsGrid[{{ContourPlot3D[dodeq == 0, {x, -9/8, 9/8}, {y, -9/8, 9/8}, {z, -9/8, 9/8}, 
                             Axes -> None, Boxed -> False, BoxRatios -> Automatic, 
                             ColorFunction -> (ColorData["TemperatureMap",
                                               LogisticSigmoid[5 gcdod[#1, #2, #3]]] &), 
                             ColorFunctionScaling -> False, Mesh -> False, 
                             PlotLabel -> "true K", PlotPoints -> 75], 
               ContourPlot3D[dodeq == 0, {x, -9/8, 9/8}, {y, -9/8, 9/8}, {z, -9/8, 9/8}, 
                             Axes -> None, Boxed -> False, BoxRatios -> Automatic, 
                             ColorFunction -> (ColorData["TemperatureMap", 
                                               LogisticSigmoid[5 mcdod[#1, #2, #3]]] &), 
                             ColorFunctionScaling -> False, Mesh -> False, 
                             PlotLabel -> "true H", PlotPoints -> 75]},
              {Graphics3D[GraphicsComplex[pts, {EdgeForm[], Polygon[tri]}, 
                                          VertexColors -> Map[ColorData["TemperatureMap"],
                                                              LogisticSigmoid[5 gc]], 
                                          VertexNormals -> nrms],
                          Boxed -> False, Lighting -> "Neutral",
                          PlotLabel -> "estimated K"], 
               Graphics3D[GraphicsComplex[pts, {EdgeForm[], Polygon[tri]}, 
                                          VertexColors -> Map[ColorData["TemperatureMap"],
                                                              LogisticSigmoid[5 mc]], 
                                          VertexNormals -> nrms],
                          Boxed -> False, Lighting -> "Neutral",
                          PlotLabel -> "estimated H"]}}]

dodecahedral surface colored by curvature


As an additional example, here is the result of using the procedure on ExampleData[{"Geometry3D", "Triceratops"}, "MeshRegion"]: triceratops colored by curvature


As noted, the literature on curvature estimation looks to be vast; see e.g. this or this or this. I'll try implementing those as well if I find time.

$\endgroup$
18
  • $\begingroup$ P.S. I had also attempted the procedure on the Stanford bunny (ExampleData[{"Geometry3D", "StanfordBunny"}, "MeshRegion"]), but the neighbor generation took too long, and failed on some of the vertices due to apparently weird geometry. If anybody wants to take a crack at it, be my guest. $\endgroup$ Jun 13, 2016 at 6:08
  • $\begingroup$ @J.M. That is great, I will also have a bit of a think about how to deal with meshes with holes. I should also check with my code for negative mean curvatures to see if they are plotting correctly. Many thanks! I hope this will also be useful to others as well. $\endgroup$
    – Dunlop
    Jun 13, 2016 at 8:42
  • 3
    $\begingroup$ Basically the same idea, but a wee bit faster: nbrs = Table[List @@@ First@FindCycle[tri ~Part~ SparseArray[Unitize[tri - v], Automatic, 1]["NonzeroPositions"][[All, 1]] /. {{a_, b_, v} :> a -> b, {b_, v, a_} :> a -> b, {v, a_, b_} :> a -> b}] // Flatten // DeleteDuplicates, {v, Length@pts}]; $\endgroup$
    – Michael E2
    Jun 13, 2016 at 12:47
  • $\begingroup$ @J.M. I have just seen that my code doesn't calculate negative curvatures and am trying to implement this now to fix things up. My code is about twice as slow as J.M.'s code, but this is perhaps because I do a nearest neighbour calculation twice. Once to calculate the vertex neighbours ('posinEdgeList'), and once to calculate which triangles are associated with each vertex ('posinTriangleList'). The second calculation could be useful to solve the edge problem though as the function 'trianglesAtLines' in my code can then be used to see which edges touch only one triangle. $\endgroup$
    – Dunlop
    Jun 13, 2016 at 12:54
  • 1
    $\begingroup$ @J.M. It is working, for a unit sphere we both now get ~4*pi as the surface area. What is also reassuring is that now I get the same curvature values as you; i.e. if I plot your Mean (or Gaussian) curvature per vertex against mine we get a nice straight line with slope 1, and I only get small differences (~10^-16) between the results (probably due to me using the built in cot function). Thanks so much for the help (I have learnt enormous amounts about mathematica as a result!). My next steps will be to check output for other surfaces, and hopefully to deal with holes/boundaries. Thanks again! $\endgroup$
    – Dunlop
    Jun 15, 2016 at 10:23
21
$\begingroup$

It took me a while, but the suggestion of @Michael E2 was quite helpful, and especially the post (Optimize inner loops).

For those of you (like me) who are new to this style of programming in Mathematica there are a few things that helped in my particular example.

In my slow version I was looping over all vertices in the mesh list.

For example in calculating the nearest neighbour list I was doing the following in the slow code:

For[i = 1, i < nvert + 1, i++,

    bb = Select[
    MeshCells[ℛ, 1][[All, 1]], #[[1]] == i || #[[2]] == i &];
    bb1 = Cases[Flatten[bb], Except[i]];

    (*Count Number of Nearest vertices on mesh*)
    nnbrs[[i]] = Length[bb1];
]

This was then replaced by:

edgeList = MeshCells[ℛ, 1][[All, 1]];
posinEdgeList = Position[edgeList, #] & /@ Range[nvert];
nearestneighbourList = Length[posinEdgeList[[#]]] & /@ Range[nvert];

which although still not blindingly fast was much faster than the previous version.

The same sort of approach was used to create functions that then could be mapped to lists of indices (for triangles, edges, vertices etc).

The faster code is now:

aellipse = 1;
bellipse = 0.7;
cellipse = 0.4;

a = BoundaryDiscretizeRegion[
  Ellipsoid[{0, 0, 0}, {aellipse, bellipse, cellipse}], 
  MaxCellMeasure -> {"Length" -> 0.15}]

ℛ = a;
(*Angles at each triangle*)
va = 
 VectorAngle[#1 - #2, #3 - #2] & @@@ Partition[#, 3, 1, {2, -2}] & /@ 
  MeshPrimitives[ℛ, {2}][[All, 1]];
(*List of labels of triangles and positions in the list at which the vertices are obtuse*)
obtusetrianglelist = 
 Position[va, n_ /; n > π/2]; 
(*Coordinates of Vertices on Mesh*)
mc = MeshCoordinates[ℛ];
(*Number of vertices*)
nvert = MeshCellCount[ℛ, 0];
(*Number of edges*)
nedge = MeshCellCount[ℛ, 1];
(*Number of faces*)
nfaces = MeshCellCount[ℛ, 2];
(*List of Edges, consisting of a list of pairs of vertex numbers*)
edgeList = MeshCells[ℛ, 1][[All, 1]];
(*List of Triangles consisting of a list of triples of vertex numbers*)
triangleList = MeshCells[ℛ, 2][[All, 1]];
(*Triangle Areas*)
areaTriangles = PropertyValue[{ℛ, 2}, MeshCellMeasure]; 
(*Length of Edges*)
edgeLengths = PropertyValue[{ℛ, 1}, MeshCellMeasure];


(*Positions of vertex i in the edge list (*SLOW*), Note this gives the edge index and either 1 or 2 depending on the order inside that edge*)
posinEdgeList = Position[edgeList, #] & /@ Range[nvert];
(*Positions of vertex i in the triangle list (*SLOW*), Note this gives the triangle index and either 1, 2 or 3 depending on the order inside that triangle *)
posinTriangleList = Position[triangleList, #] & /@ Range[nvert];

(*Number of nearest neighbours at each vertex*)
nearestneighbourList = Length[posinEdgeList[[#]]] & /@ Range[nvert];

(*Function that calculates for a given pair of vertex indices from a line i,j, what triangles in the mesh indices also contain these indices, output is the triangle index*)

trilistfunc[line_] := Intersection[posinTriangleList[[line[[1]]]][[All, 1]],posinTriangleList[[line[[2]]]][[All, 1]]];
(*List of triangle indices that are attached to each line, This means that trianglesAtLines[[k]] will return the indices of the triangles containing the line k (If only one index is returned we are on the boundary!)*)
trianglesAtLines = Map[trilistfunc, edgeList];
(*Function that calculates which vertices in the attached triangles to a given edge are opposite to this edge, vertices are given as indices*)

(*List of indices of edges that are on the boundary*)
boundaryedges = Flatten[Position[Length[trianglesAtLines[[#]]] & /@ Range[nedge], 1]];
(*List of indices of vertices that are on the boundary*)
boundaryvertices = Flatten[edgeList[[#]] & /@ boundaryedges] // DeleteDuplicates;

(*Function that calculates which vertices in the attached triangles to a given edge are opposite to this edge, vertices are given as indices*)
oppcornerfunction[i_] := If[MemberQ[boundaryedges, i], {0, 0},{Cases[Cases[triangleList[[trianglesAtLines[[i, 1]]]], Except[edgeList[[i, 1]]]], Except[edgeList[[i, 2]]]][[1]], Cases[Cases[triangleList[[trianglesAtLines[[i, 2]]]], Except[edgeList[[i, 1]]]], Except[edgeList[[i, 2]]]][[1]]}];

(*List of pairs of vertex indices m and n, that are opposite to edge k, pairs are ordered according to the edge number, if {0,0} then on boundary*)
oppcornerList = Map[oppcornerfunction, Range[nedge]];

(*Function that calculates the cotangents of the angles of the corners of "both" triangles opposite to a given edge ordered according to edge number (gives 0,0 for edge)*)

cotanglepairfunc[i_] := If[MemberQ[boundaryedges, i], {0, 0}, {Cot[va[[trianglesAtLines[[i, 1]]]][[Position[triangleList[[trianglesAtLines[[i, 1]]]], oppcornerList[[i, 1]]][[1, 1]]]]], Cot[va[[trianglesAtLines[[i, 2]]]][[Position[triangleList[[trianglesAtLines[[i, 2]]]], oppcornerList[[i, 2]]][[1, 1]]]]]}];
(*List of pairs of the cotangents of the opposite corner angles to each edge k*)
canglepairs = Map[cotanglepairfunc, Range[nedge]];

(*Function so we choose vertex j attached to vertex i*)

sw12func[a_] := If[a[[2]] == 1, 2, 1];
(*Function to calculate the list of oriented ij vectors attached to vertex i*)
ijvectfunc[i_] := Map[(mc[[i]] - mc[[edgeList[[posinEdgeList[[i, #, 1]],sw12func[posinEdgeList[[i, #]]]]]]]) &, Range[Length[posinEdgeList[[i]]]]];
(*List of oriented ij vectors attached to each vertex i *)
ijvectlist = Map[ijvectfunc, Range[nvert]];
(*Function to calculate the Mean curvature vector at each vertex*)

ijCVfunc[i_] := Total[ijvectlist[[i]]*Map[Total[canglepairs[[posinEdgeList[[i, #, 1]]]]] &, Range[Length[posinEdgeList[[i]]]]]];
(*List of Mean Curvature vectors at each vertex*)
ijCVlist = Map[ijCVfunc, Range[nvert]];

(*Now we need to calculate the Voronoi Area, modified such that obtuse triangles are taken into account*)
(*Create Function to \
calculate mixed Voronoi Area (see paper for explanation)*)

trianglecoords = Map[mc[[#]] &@triangleList[[#]] &, Range[nfaces]];
faceNormalfunc[tricoords_] := Cross[tricoords[[1]] - tricoords[[2]], 
tricoords[[3]] - tricoords[[2]]];
facenormals = Map[faceNormalfunc, trianglecoords];
mcnewcalc = Map[Total[Map[(facenormals[[#]]*areaTriangles[[#]]) &,posinTriangleList[[All, All, 1]][[#]]]] &, Range[nvert]];
meancvnew = -Sign[MapThread[Dot, {mcnewcalc, ijCVlist}]] (Norm /@ ijCVlist);

(*Now we need to calculate the Voronoi Area, modified such that obtuse triangles are taken into account*)
(*Create Function to calculate mixed Voronoi Area (see paper for explanation)*)

areaMixedfunction[i_] := If[MemberQ[boundaryvertices, i], 0,Total[Map[Do[

     edgenumber = posinEdgeList[[i, #, 1]];

     d1 = trianglesAtLines[[edgenumber]][[1]];
     d2 = trianglesAtLines[[edgenumber]][[2]];
     AMixed = 0;
     If[MemberQ[obtusetrianglelist[[All, 1]], d1],
      (*Now do test to see which triangle area we add correcting for whether the triangles are obtuse or not*)

      ObtVnum = Position[obtusetrianglelist[[All, 1]], d1][[1, 1]];

      (*Vertex index of the obtuse part of the triangle*)

      Vnum = triangleList[[obtusetrianglelist[[ObtVnum, 1]],obtusetrianglelist[[ObtVnum, 2]]]];
      If[Vnum == i,
       (*Triangle Obtuse at i, therefore add half of area T/2*)

        AMixed += (1/4)*areaTriangles[[d1]];
       ,
       (*Triangle Obtuse but not at i, therefore add half of area T/4*)

              AMixed += (1/8)*areaTriangles[[d1]];
       ]
      ,
      AMixed += (1/8)*(canglepairs[[edgenumber]][[1]])*(edgeLengths[[edgenumber]])^2
      (*If False we add the normal voronoi*)
      ];
     (*Repeat the test for the other triangle*)

     If[MemberQ[obtusetrianglelist[[All, 1]], d2],
      (*Now do test to see which triangle area we add*)

      ObtVnum = Position[obtusetrianglelist[[All, 1]], d2][[1, 1]];
      Vnum = 
       triangleList[[obtusetrianglelist[[ObtVnum, 1]], 
         obtusetrianglelist[[ObtVnum, 2]]]];;
      If[Vnum == i,
       (*Triangle Obtuse at i, therefore add half of area T/2*)

        AMixed += (1/4)*areaTriangles[[d2]];
       ,
       (*Triangle Obtuse but not at i, therefore add half of area T/4*)

              AMixed += (1/8)*areaTriangles[[d2]];
       ]
      ,
      AMixed += (1/8)*(canglepairs[[edgenumber]][[2]])*(edgeLengths[[edgenumber]])^2
      (*If False we add the normal voronoi*)
      ];
     Return[AMixed]
     , 1] &, Range[Length[posinEdgeList[[i]]]]]]];
(*Create a list of the Mixed areas per vertex*)
AmixList = Map[areaMixedfunction, Range[nvert]];
(*Gaussian Curvature*)
K = If[MemberQ[boundaryvertices, #], 0,Map[(2*π - Total[Extract[va, Position[MeshCells[ℛ, 2][[All, 1]], #]]])/AmixList[[#]] &, Range[nvert]]];
H = If[MemberQ[boundaryvertices, #], 0,Map[(meancvnew[[#]]/AmixList[[#]])/4 &, Range[nvert]]];

Now this code doesn't yet deal with surfaces with boundaries or holes (One probably needs to exclude edges from the calculation), so only works with closed surfaces. Still not blindingly fast but much better.

An example of how this code can be used is below:

Image of ellipses coloured by mean and gaussian curvature

Or on the surface defined by:

a = BoundaryDiscretizeRegion[
  ImplicitRegion[
   x^6 - 5 x^4 y z + 3 x^4 y^2 + 10 x^2 y^3 z + 3 x^2 y^4 - y^5 z + 
     y^6 + z^6 <= 1, {x, y, z}], MaxCellMeasure -> {"Length" -> 0.2}]

Other interesting surface

What I am not sure about is how accurate this estimation is, which I can now test systematically now the code is faster. The paper cited also allows the calculation of the principle curvature directions, but this requires solving for the local curvature tensor at each vertex. I hope to get around to this as well, but perhaps this should be another post on another topic?

Thanks for the input @Michael E2, this was the key to speeding things up, and I hope this code could be useful.

EDIT:

The plotting code I used is as follows:

To Calculate ellipse curvatures I used the equations from MathWorld for ellipse curvatures. I am not entirely sure whether my implementation is correct however.

Vertexangles = Table[{}, {ii, 1, nvert}];
Kellipse = Table[{}, {ii, 1, nvert}];
Hellipse = Table[{}, {ii, 1, nvert}];
(* Analytical Curvature Calculations from Mathworld: \
http://mathworld.wolfram.com/Ellipsoid.html *)
For[i = 1, 
 i < nvert + 1, i++,
 (*We calculate here the angles of each of the vertices, equivalent \
to uv coordinates in the mathworld ellipse definition*)

 Vertexangles[[
   i]] = {VectorAngle[{mc[[i, 1]], mc[[i, 2]], 0}, {1, 0, 0}], π/
     2 + VectorAngle[mc[[i]], {mc[[i, 1]], mc[[i, 2]], 0}]};
 (*Analytical Gaussian Curvature calculation as a function of angle, \
parameterisation*)

 Kellipse[[i]] = (
  aellipse^2 bellipse^2 cellipse^2)/(aellipse^2 bellipse^2 Cos[
      Vertexangles[[i, 2]]]^2 + 
    cellipse^2 (bellipse^2 Cos[Vertexangles[[i, 1]]]^2 + 
       aellipse^2 Sin[Vertexangles[[i, 1]]]^2) Sin[
      Vertexangles[[i, 2]]]^2)^2;
 (*Analytical Mean Curvature calculation as a function of angle, \
parameterisation*)

 Hellipse[[
   i]] = (aellipse bellipse cellipse (3 (aellipse^2 + bellipse^2) + 
       2 cellipse^2 + (aellipse^2 + bellipse^2 - 2 cellipse^2) Cos[
         2*Vertexangles[[i, 2]]] - 
       2 (aellipse^2 - bellipse^2) Cos[2*Vertexangles[[i, 1]]]*
        Sin[Vertexangles[[i, 2]]]^2))/(8 (aellipse^2 bellipse^2 Cos[
         Vertexangles[[i, 2]]]^2 + 
       cellipse^2 (bellipse^2 Cos[Vertexangles[[i, 1]]]^2 + 
          aellipse^2 Sin[Vertexangles[[i, 1]]]^2) Sin[
         Vertexangles[[i, 2]]]^2)^(3/2));
 ]

Then to plot I used

Kmin = 0;
Kmax = 10;
Hmin = 0;
Hmax = 10;
Hellrescale = Rescale[Hellipse, {Hmin, Hmax}];
Kellrescale = Rescale[Kellipse, {Kmin, Kmax}];
Hrescale = Rescale[H, {Hmin, Hmax}];
Krescale = Rescale[K, {Kmin, Kmax}];
v = MeshCoordinates[ℛ];
w = MeshCells[ℛ, 2];
discreteMCgraph = 
  Graphics3D[{EdgeForm[], 
    GraphicsComplex[v, w, 
     VertexColors -> 
      Table[ColorData["TemperatureMap"][Hrescale[[i]]], {i, 1, 
        nvert}]]}, Boxed -> False, PlotLabel -> "Discrete"];
exactMCgraph = 
  Graphics3D[{EdgeForm[], 
    GraphicsComplex[v, w, 
     VertexColors -> 
      Table[ColorData["TemperatureMap"][Hellrescale[[i]]], {i, 1, 
        nvert}]]}, Boxed -> False, PlotLabel -> "Analytical"];
discreteGCgraph = 
  Graphics3D[{EdgeForm[], 
    GraphicsComplex[v, w, 
     VertexColors -> 
      Table[ColorData["TemperatureMap"][Krescale[[i]]], {i, 1, 
        nvert}]]}, Boxed -> False, PlotLabel -> "Discrete"];
exactGCgraph = 
  Graphics3D[{EdgeForm[], 
    GraphicsComplex[v, w, 
     VertexColors -> 
      Table[ColorData["TemperatureMap"][Kellrescale[[i]]], {i, 1, 
        nvert}]]}, Boxed -> False, PlotLabel -> "Analytical"];

Legended[GraphicsGrid[{{discreteMCgraph, exactMCgraph}}, 
  PlotLabel -> "Mean Curvature"], 
 BarLegend[{"TemperatureMap", {Hmin, Hmax}}]]
Legended[GraphicsGrid[{{discreteGCgraph, exactGCgraph}}, 
  PlotLabel -> "Gaussian Curvature"], 
 BarLegend[{"TemperatureMap", {Kmin, Kmax}}]]

(Certainly more work to be done, but hopefully this is useful to others as well)

EDIT 2: The code has now been edited to more carefully account for negative mean curvatures. I used a slightly different method to @J.M. to estimate vertex normals (I calculated face normals and then averaged them, weighted by the face area). I then used the same line of code of J.M. used to work out whether this is on the same side of the surface as the "curvature vector" hence allowing for the calculation of the sign.

EDIT 3: I noticed in the last calculation of the Mean Curvature that I had missed the statement in the MDSP paper that shows the mean curvature value is half the magnitude of equation (8), this is now corrected in the code).

EDIT 4: The surface now deals with holes, not the most elegant solution, but seems to work well, and seems to give a good agreement between analytical and estimated values of mean and gaussian curvatures. See examples below.

A constant mean curvature surface Catenoid (with holes) showing estimated mean and gaussian curvatures:

Catenoid Curvatures

Images of curvatures (Mean and Gaussian) on a meshed unit sphere

Unit Sphere

Plots of analytical vs exact curvature

Analytical vs Estimated Curvatures

Images of curvatures (Mean and Gaussian) on a meshed ellipsoid (two major axes with length 1 and one axis with length 0.7)

Ellipsoid 1,1,0.7

Comparison of Exact and Estimated Mean and Gaussian Curvatures for this ellipsoid

Exact vs Estimated Curvatures Ellipsoid 1,1,0.7

Images of curvatures (Mean and Gaussian) on a meshed ellipsoid (One major axis with length 1, one with length 0.7 and one with length 0.3)

Ellipsoid 1,0.7,0.3

Comparison of Exact and Estimated Mean and Gaussian Curvatures for this ellipsoid

Comparison Exact vs Analytical Ellipsoid 1,0.7,0.3

As can be seen the results are somewhat noisy, but pretty consistent. The implementation of the "hole" or "boundary" finding is a bit clunky (and is my code in comparison to @ J.M.s) and there must be much more elegant ways of doing this. For my purposes this is certainly useable and gives a lot of ideas of how to improve in the future. I hope this is helpful to others as well.

EDIT 5: The code is now correctly plots negative mean curvatures. Many thanks to Dr Volkov for letting me know that it wasn’t working correctly. In the end there was an extra Norm in the final equation which is now removed.

Plotted here are images of “erythrocyte-like” shapes plotted using the following equations taken from the paper: Larkin, T.J. & Kuchel, P.W. Bull. Math. Biol. (2010) 72: 1323. Mathematical Models of Naturally “Morphed” Human Erythrocytes: Stomatocytes and Echinocytes. These shapes both have positive and negative mean curvature.

Erythrocyte shapes were calculated using the following code:

P[m_, d_] := (1 - 2 m) d^2/(2 m );
Q[m_, d_, a_] := (1 - m) d^4/16 a^2 m;
R[m_, d_] := (m - 1) d^4/16 m;
eryth[m_, d_, a_] := (x^2 + y^2)^2 + P[m, d] (x^2 + y^2) + 
   Q[m, d, a] z^2 + R[m, d];

Images of erythrocyte-like shapes showing exact and discrete mean and gaussian curvatures

$\endgroup$
1
  • 3
    $\begingroup$ For completeness, can you please include the plotting code that you used as well? $\endgroup$ Jun 10, 2016 at 5:02
5
$\begingroup$

This is not an answer but a comment on the speed issue @J.M. mentioned:

One thing that can be done is to use

Needs["NDSolve`FEM`"]
bmesh = ToBoundaryMesh[r, {{-1.1, 1.1}, {-1.1, 1.1}, {-1., 1.}}, 
   MaxCellMeasure -> {"Length" -> 0.1}];
(* bmesh = ToBoundaryMesh[dod] *)

pts = bmesh["Coordinates"];
tri = Join @@ ElementIncidents[bmesh["BoundaryElements"]];

You can then:

AbsoluteTiming[
 bmesh["VertexBoundaryConnectivity"];
 temp = Table[
   DeleteDuplicates[Join @@ Extract[tri, k["NonzeroPositions"]]], {k, 
    bmesh["VertexBoundaryConnectivity"]}];
 nbrs2 = (DeleteDuplicates /@ 
     Join[Table[{k}, {k, Length[temp]}], temp, 2])[[All, 2 ;; -1]];
 ]

(* {0.03484, Null} *)

This is about an order of magnitude faster.

Note that nbrs2 is not exactly nbrs but this holds:

Sort[Sort /@ nbrs] === Sort[Sort /@ nbrs2]
True

Ideally, however, one would use the temp result above and not nbrs2.

$\endgroup$
1
  • $\begingroup$ Thanks, I'll see how to best use this method for the neighbors. As I had noted, a reliable method ought to produce the neighbors in anticlockwise order as long as all the faces of the mesh are oriented anticlockwise. $\endgroup$ Jun 20, 2016 at 1:18
0
$\begingroup$

Iterate through triangles other than vertex, then you don't need to find neighbors.

(Initial:)

MeanC = Table[{0, 0, 0}, Length[vlist]];
Avor = Table[0, Length[vlist]];
GauC = Table[2 \[Pi], Length[vlist]];

(Loop)

For[ti = 1, ti <= Length[tlist], ti++,
 For[i = 1, i <= 3, i++,
  vi = tlist[[ti, i]];
  {j, k} = Delete[{1, 2, 3}, i];
  {vj, vk} = tlist[[ti]][[{j, k}]];  
  MeanC[[vi]] += 
   1/2 Cot[va[[ti, k]]] (vlist[[vi]] - vlist[[vj]]) + 
    1/2 Cot[va[[ti, j]]] (vlist[[vi]] - vlist[[vk]]);
  Avor[[vi]] += 
   If[va[[ti, 1]] > 0.5 \[Pi] || va[[ti, 2]] > 0.5 \[Pi] || 
     va[[ti, 3]] > 0.5 \[Pi],
    If[va[[ti, i]] > 0.5 \[Pi], 1/2 Alist[[ti]], 1/4 Alist[[ti]]], 
    1/8 Cot[va[[ti, k]]] Norm[vlist[[vi]] - vlist[[vj]]]^2 + 
     1/8 Cot[va[[ti, j]]] Norm[vlist[[vi]] - vlist[[vk]]]^2];;
  GauC[[vi]] += -va[[ti, i]]]
 ]

(Finalize)

h = Table[1/2 Norm[MeanC[[i]]/Avor[[i]]], {i, 1, Length[vlist]}];
k = Table[GauC[[i]]/Avor[[i]], {i, 1, Length[vlist]}];
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.