Handling Large Shrinking Deltas for Polygons

[bektaskemal]

Hi, and thanks for the excellent work on this library!

I want to use this library for shrinking polygons and it worked well for me during tests. The operation understandably fails when the delta exceeds size limits, but for my specific use case, I need a slightly different behavior.

I've attached an image illustrating the desired behavior:

For example, with a rectangle, I would like the following behavior:
delta < w/2: The offset behaves normally.
w/2 <= delta < h/2: The shape reduces to a line in the center, which becomes smaller as the delta increases.
h/2 <= delta: The shape reduces to a single point in the center.

I'm aiming for similar behavior with more complex shapes, as shown in the image.

Would it be possible to achieve this behavior with the current library? If not, do you have any suggestions or workarounds that might help?

Thanks in advance for your assistance!

[AngusJohnson]

Hi Kernal.
I don't think what you're requesting would have such wide appeal that it would be worth my coding effort.
But you have all the library's source code, so you can always customise it to your own specific requirements.

[niki-sp]

Is it possible to approximate max delta for non-empty result?

[AngusJohnson]

Is it possible to approximate max delta for non-empty result?

Not really. At least, not without very significantly impairing performance.

[niki-sp]

Sorry for ambiguous question. I will try again:
Is it possible to approximate max delta as new function (not in existing code)?

[reunanen]

At the risk of stating the obvious... If you are not too worried about performance, I think you could try to do something like this:

First estimate the maximum possible max delta. Guessing you could for example calculate sqrt(w*w+h*h) (where w and h are the width and height of the bounding rectangle), and divide the result by 2 perhaps.

Now you have limits for the minimum (0) and maximum max delta. At the low limit, the offsetting result should be non-empty, and at the high limit it should be empty. Now you should be able to do a binary search (or similar) to narrow down this range, and in particular find the largest delta that just doesn't make the offsetting result be empty. By increasing the number of search steps, you can reach arbitrarily high precision, I think.

Does that make sense?

Yep, I fully understand that you may be looking for something that works much faster than this iterative approach.

[niki-sp]

I was hoping for something like R of the max nested circle

[reunanen]

Ok, so something like this maybe?
https://github.com/mapbox/polylabel