Dividing A Polygon In Any Given Number Of Equal Areas

This post will detail a closed form solution to split polygons into any number of equal areas, while ensuring minimum length of line based cuts. The solution works for both convex and concave polygons, as long as they don't have non-manifold vertices, and can be traversed with a single loop. Credit for this concept goes to Dr. Rossignac, who also introduced me to the power of medial axis.

Figure 1: Examples of Convex (Left) and Concave (Right) Polygon division. Red lines show the cuts splitting polygons into equal areas

Base Case: 4-Sided Polygon
Let's start with a 4-sided polygon, and say we need to split this into N sub-polygons. We use the divide and conquer methodology:

  • Calculate the area of the polygon, say A_{poly}. Therefore the target area of each of the N sub-polygons would need to be A_{poly} \over N
  • Split the polygon into two sub-polygons - smaller one with area of A_{poly} \over N, and bigger one with area of {(N-1)\over N} A_{poly}
  • Continue using the same approach on the bigger polygon till the bigger polygon reduces into polygon with desired area.

The problem of splitting polygon N-ways has now been reduced to splitting polygon into two sub-polygon with specific area. Since the cut is a line, it would start from one edge and end at another. There are six possible edge pairs (^4 C_2) in a 4-sided polygon, that could lead to a potential cut. Three of these cuts - c_1, c_2, c_3 are shown in the images below as a reference. We can select the minimum cut from the all potential cuts obtained from each edge pair.

Figure 2: Three potential cuts for splitting polygon ABCD

Now, the problem is reduced to making a cut on a given edge pair, such that the cut splits the polygon into two sub-polygons of specific area.

Splitting Polygon For a Given Edge Pair
Say, we select edges AD, and BC in the 4-sided polygon shown below.

  • Determine the angle bisector of the edge pair. This is dotted line, labelled as m in the figure below.
  • Determine the projected points of the four vertices on their corresponding opposing edge, at the angle perpendicular to the angle-bisector m
  • Ignore the projected points that don't lie on the opposing line segment. In the figure, the projected vertex of B is G, and D is H. The projection of point A along the line perpendicular to angle-bisector m doesn't lie on edge BC, so it is ignored. Same argument applies to the projection of point C. The valid projection points of the four vertices are G, and H.
  • Connect the vertices B, and D to their corresponding valid projection points G, and H. The polygon formed by the edge pair, is now a combination of triangles AGB, DCH, and trapezoid GDHB

    Figure 3: Splitting polygon for edge pair: AD and BC

  • The minimum cut, if it exists for this given edge pair, must lie in these two triangles or the trapezoid
  • To find out where the minimum cut line lies, we can do a simple area check. Say, A_{AGB} is the area of the left triangle, A_{GDHB} is area of the trapezoid, and A_{DCH} is the area of the right triangle. The minimum cut line lies in the:
    • Left Triangle, if {A_{poly} \over N} < A_{AGB}
    • Trapezoid, if {A_{poly} \over N} > A_{AGB}, and {A_{poly} \over N} < A_{AGB} + A_{GDHB}
    • Right Triangle, if {A_{poly} \over N} < A_{AGB} + A_{GDHB}
  • Minimum Cut in a Triangle For a Given Edge Pair
    Say, based on the area test, the minimum cut lies inside the left triangle, as shown in Figure 4. One end of the red cut line lies on the vertex B. The other end of the edge can be found by linear interpolation of points A, and G based on the target area. Since the target area is A_{poly} \over N, and the interpolated point will be A + {A_{poly} \over N*A_{AGB}}(G - A)

    Figure 4: Minimum cut (Red Line), for a given edge pair lies in the triangle AGB

  • Minimum Cut in a Trapezoid For a Given Edge Pair
    If the minimum cut lies in the trapezoid, then the two points of the split line (shown in red) can be found via linear interpolation based on the target area, similar to the approach we took for the triangle case above. In the case of triangle, one point of the split line was fixed as a vertex. In the case of trapezoid, both points of the red split line can move based on the area. However, the requirement that the split line needs to be minimum dictates that the angle of the split line must always be perpendicular to the angle-bisector of the edge pair. One end of the line will be the linear interpolation between points G, and D. The other end of the red line will be the linear interpolation between points B, and H.

    Figure 5: Minimum cut (Red Line), for a given edge pair lies in the trapezoid

Key Concept
For a given edge pair, the minimum cut line within a trapezoid, must lie perpendicular to the angle-bisector of the two edges in the edge-pair. While this makes sense intuitively, it can also be geometrically proved by creating a trapezoid where the parallel edges lie at an angle \alpha to the angle-bisector. The area of this trapezoid can be expressed as a function of \alpha. For the cut line to be minimum, we take the derivative of the area, and it can be shown, that for a given area, \alpha must be equal to 90^{\circ}

General Case: Splitting N-Sided Polygon
So far, we have shown, how to determine the minimum cut line for a given edge pair. We can easily extend this for any N-sided polygon

  • For a given edge, we check all the other remaining edges, to see where a split line could exist
  • The figure above shows a random polygon, and we have selected an edge pair AB, and CD
  • There may be several edge pairs that provide a potential split cut. We can compare them, and select the minimum cut.
  • Special Case: For concave polygons, some of the cut lines found with above approach will be outside the polygon and/or intersect with existing edges. So we simply add an additional verification step to ensure that the split lines don't do so. If they do, then we ignore those split lines as a potential solution.

Figure 6: General case: splitting n-sided polygon

Time Complexity
Say, we want to split a 4-sided polygon into 4 sub-polygons. We start by splitting the polygon into two pieces one with {1/4}^{th} area, and another one with {3/4}^{th} area. The same process would then be repeated on the bigger sub-polygon till all sub-polygons are of target area. At each step of splitting into two sub-polygons, there are several possible edge pairs. For example, in the very first step, the 4-sided polygon has 6 possible edge pairs, and therefore 6 possible split lines. Each of these splits would lead to its own tree of final solution. Therefore we have two possible approaches:

  • Greedy & Iterative Approach
    We take the minimum cut at each step, and ignore the others. This approach may not always lead to the correct solution, but the results will be fast. The algorithm has cubic time complexity for convex polygons. Checking every edge pair requires quadratic complexity, and for a given edge pair, the algorithm involves calculating area, and doing linear interpolation for finding split lines. For concave polygons, another verification step, that involves ensuring split lines don't intersect with other remaining edges, is needed. This additional step leads to quartic time complexity.
  • Comprehensive & Recursive Approach
    In this approach, we consider every possible split generated by each edge pair at each step, and follow the trail from each split. Because of the recursive nature, the result is always correct, but the time complexity becomes exponential.

The recursive approach as outlined above was implemented via processing. The video demo below shows how the split lines change in real time as polygon coordinates or number of required splits is modified.

  • James

    Great post and thanks for sharing. I would like to try and implement your solution in javascript, unfortunately the maths is a bit beyond me, any chance you could share your processing code as a starting point, many thanks.

  • M. T

    Brilliant algorithm! I have a question,
    In your n-sided polygon case, the angle-bisector had already intersected with other existing edges, as the red mark.
    Does that mean I should ignore this angle-bisector, giving up the edge pair(AB&CD) and try another edge pair?

    • http://www.twitter.com/khetarpal Khetarpal

      MT, Glad you found the post useful. No need to check whether the angle bisector intersects with other edges. It is the minimum cut line that shouldn't intersect with other edges. For a given edge pair, the minimum cut line will either be in the trapezoid (shown in blue) or the triangles (shown in pink). Hope this helps.

  • Erik

    Great work! For my thesis I'm currently working on the dividing a polygon in known areas (for example you have as input with known order A: 10 m2, B: 14 m2, C: 8 m2). Do you think it is possible to adjust this algorithm to make that possible?

    • http://www.twitter.com/khetarpal Khetarpal

      Erik, certainly! that would require only a minor tweak to the algorithm. Instead of dividing into equal areas, just specify the known areas, and you should be good to go.

      • Erik

        That would be very nice! Do you have a github where I could find the code and maybe a paper so I could cite you?

  • Gediminas Rimša

    Hello and thanks for sharing.
    I'm working on an implementation of this algorithm for JTS (Java Topology Suite) polygons and would like to clarify a few things:
    1) How would you determine projection points for an edge pair which would intersect if extended? Especially when edges are perpendicular to each other, e.g. vertical edge where X = 0 to 20 and horizontal edge where Y = 5 to 15.
    2) How should we handle cases when some part of e.g. Triangle 1 falls outside of the full polygon being split? Continue calculation based on real (reduced) area or just discard the edge pair?

    The current implementation we have is available at: https://github.com/grimsa/polysplit

    • http://www.twitter.com/khetarpal Khetarpal

      Hi Gediminas!

      1a) It is okay for the edges to intersect if they are extended. In-fact you will notice that in above Figures (2 & 3), the edge pair (edges AD & BC) does intersect when extended.

      1b) For the specific case that you mention, the minimum cut line won't go through those two edges (X = 0-20 & Y=5-15). While it is certainly possible to split the polygon through those two edges, it won't be the minimum possible length of cut. My solution only covers the *minimum length* of cut. For your example, such a minimum length cut may come from other edge pairs.

      2. Re: your point # 2, it is helpful to have some some specific test cases. I would recommend start off with convex polygons first, and when that is working, move to the concave polygons later. The scenario you are mentioning will happen for concave polygons - some of the cut lines found may be outside the polygon and/or intersect with existing edges. So we simply add an additional verification step to ensure that the split lines don't do so. If they do, then we ignore those split lines as a potential solution.

      • Gediminas Rimša

        I've drawn up some pictures - will address each case in a separate post :)

        1a) Sorry, I was not clear enough. I meant the case when the intersection point falls on one of the edges.
        The way I have it implemented currently is displayed in the image. First of all, I create a line 'p' passing through the intersection point X and that is perpendicular to the edge on which we are trying to project the point.
        Then the logic is the following:
        if (intersection point is on the edge we are projecting onto) {
        Then we shorten the edge by replacing one of the vertexes by intersection point and proceed as usual
        } else {
        If point and edge are on the opposite sides of line p, projection will fall outside of the edge.
        Maybe there is some smarter way to do this?

      • http://www.twitter.com/khetarpal Khetarpal

        The algorithm I described first calculates the angle bisector of the edge pair, and then the line p is perpendicular to the angle bisector. Your line 'p' is perpendicular to the line where you are taking the projection. This will not lead to optimal results. Is there a reason, you don't want to take 'p' as perpendicular to the angle bisector of the edge pair?

      • Gediminas Rimša

        1b) I think it is possible to construct such a case that the minimum length cut will belong to the pair of perpendicular edges. Maybe the one in the image would be it? If we have a big square (say 100x100) with an extension of a small rectangle (say, of area of 10) and we have to split the polygon into parts of size 15.

        A good thing is that I think I have an idea how to handle projections for perpendicular edges. For that we have to consider the orientation of both edges. For example, if we are traversing along the ring of the polygon, the bounded are will always lie on the left of each edge (or right, depending on direction of traversal). Let's say we are traversing in the direction of ABCDEF..., then polygon's area is on the left of each edge.
        Now if we have a pair of perpendicular edges, we can take the direction of both of them (left+left) and that will give us a quadrant. E.g. for AB and FG it would be the AFG one, for AB and DE it would be BED. Then the angle bisector we are interested in will be in that quadrant. And the rest is same as the 1a case.

      • Gediminas Rimša

        2) Yeah, if you only had to support convex polygons, it would be much easier :) Attaching a test case.
        The idea is that the cut line itself does not necessarily intersect any of the edges or fall outside of the polygon. But parts of the triangles (or even a part of trapezoid) could (e.g. XDE, marked in red).

      • http://www.twitter.com/khetarpal Khetarpal

        This test case shouldn't be a problem at all with the above algorithm. All XDE does is lead to a negative area, but that is already factored in when you calculate the area of the entire polygon to begin with.

        For purpose of explanation, I'm going to call the weird jagged polygon shape on the left of AF as J.

        The min cut line is not just splitting the polygon ABEF in your picture, but it splits the entire polygon, ABCXDEFJA . So the cut line leads to two sub-polygons. One on the left which is A(Cut Line)FJA. and another on the right which is BCXDE(CutLine)B. So, when you calculate the two areas of these two polygons, the negative area of XDE doesn't have any effect.

        Its hard to talk in hypotheticals, and explain them. I suggest make a real test case with actual dimensions, and see what min cut line you get a result of this algorithm. Keep the test case to the simplest while keeping the negative area of XDE in there.

  • Viktor Grabarchuk

    If someone is searching for solution on C++, here it is https://github.com/dhmhd/poly-split
    Thanks for sharing, Sumit Khetarpal.

    • http://www.twitter.com/khetarpal Khetarpal

      That's great Viktor. Thanks for the acknowledement.

    • JeongYong Park

      I was ported to javascript. but has some little bug.

  • Yinan Li

    Hi, can you help me with the interpolated point formula? What does A and G represent? The coordinates of A and G?

  • Wang Xin

    Thanks for sharing such good topic, and i have some questions confusing me.In my understanding, such an algorithm is a greedy algorithm that it gradually divide the scene into N parts, and in each iteration, only 1/N part is divided. How good is the partition result, especially compared to a global optimal partition? for example, when the polygon is square,i don't think the cut lines is shortest with greedy algorithm.