How make endpoints of bwboundaries consistent in Matlab?

I have an image with 2 edges

if I then plot the boundary of the edges with the code below:

imshow(I); hold on; [B,L,N] = bwboundaries(I);
for k=1:length(B),
    boundary = B{k};
    BL=size(boundary);
    plot(boundary(1,2), boundary(1,1), '*g','MarkerSize',15);
    for j=1:10:BL(1)
        plot(boundary(j,2), boundary(j,1), '.r','MarkerSize',5);
    end

end

As seen in the image above, the starting point (the green star) for the left edge is at the left side of the image, which is what I expected. However, the starting point for the right edge is towards the middle

Apparently this is because bwboundaries deals with tracing objects in clockwise direction, whereas the 2nd edge needs to be traced counterclockwise for it to begin and end on the right boundary of the image

How can Matlab be able to take the positions from bwboundaries and correctly determine the endpoints for the edge on the right?

Solution

Not a complete answer to your problem, but an idea I came up with. You can check all points of a boundary for their "closeness" to an image border, and then find minimum/maximum (x, y) values, that'll describe those "ends" you're interested in.

B = bwboundaries(img);

% Threshold for "closeness" to an image border.
thrNear = 5;

for k = 1:numel(B)

  b = B{k};

  nearTop = b(:, 1) < thrNear;
  nearBottom = b(:, 1) > (size(img, 1) - thrNear);
  nearLeft = b(:, 2) < thrNear;
  nearRight = b(:, 2) > (size(img, 2) - thrNear);

  closeToTop = b(nearTop, :)
  closeToBottom = b(nearBottom, :)
  closeToLeft = b(nearLeft, :)
  closeToRight = b(nearRight, :)

end

For example, for the right shape in your original image, you get:

closeToTop = [](0x2)
closeToBottom = [](0x2)
closeToLeft = [](0x2)
closeToRight =

    79   283
    79   284
    79   285
    79   286
    79   287
    80   287
    81   287
    81   286
    81   285
    81   284
    81   283
   215   283
   215   284
   215   285
   215   286
   215   287
   216   287
   217   287
   217   286
   217   285
   217   284
   217   283

Now, go for the maximum x value(s) (287), and find appropriate (non-neighbouring) y values (79-81 vs. 215-217). Repeat that for each image border.

I hope, you get my idea. To be honest, I don't want to implement that entirely, but don't hesitate to ask, if my description is not precise enough.