Have you ever worked with 2D boundary shape representations and noticed that the same shape can be represented with reasonable accuracy by a much smaller number of points? Did you ever need to reduce the complexity of a 2D shape while retaining as many salient features as possible? If so then DecimatePoly is the function you have been looking for.
In the attached .ZIP folder you will find the primary, self-contained function titled DecimatePoly. Additionally, in the Demos folder you will find three files demonstrating the capability and utility of this function.
DecimatePoly_demo1: Simplify a 2D contour extracted from a binary image. Three binary images are included as examples. This demo requires Image Processing Toolbox to run.
DecimatePoly_demo2: Simplifying complex shapes while retaining geometrically and visually salient features.
DecimatePoly_demo3: Use `DecimatePoly.m` to improve the run-time of in-polygon tests at the cost of minor misclassification errors.
Anton Semechko (2020). DecimatePoly (https://github.com/AntonSemechko/DecimatePoly), GitHub. Retrieved .
A Python implementation of this function may be found here:
Besides the small issue I've mentioned in my previous comments (which has been corrected in the translation), this is a direct translation from MATLAB to Python.
EDIT to my previous comment:
"Here, `t` is an N-by-1 array. `t<0` is only true when all elements of `t` are less than *zero*, so negative `t` values may pass. Similarly, `t` values greater than one may pass."
On lines 79 and 80 of DecimatePoly.m, I believe the following statements don't work as intended:
if t<0, t(t<0)=0; end %#ok<*BDSCI>
if t>1, t(t>1)=1; end
Here, `t` is an N-by-1 array. `t<0` is only true when all elements of `t` are less than one, so negative `t` values may pass. Similarly, `t` values greater than one may pass.
From the context of the two statements and the fact that we're calculating distances, I believe the author means to "clip" the values of `t` to the range (0, 1). If this is true, the conditionals should be taken out. Those two lines would be replaced as:
If I'm correct that this is an error, lines 175 and 176 display the same flawed logic. However, `t` is a 1-by-2 array within the `RecomputeErrors` function, so the statements are not erroneous but only highly redundant.
Very useful function. I would enjoy an option on deciding if you could actually save output statistics on variables instead of directly printing them (useful when I use it inside other routines).
Great function. It's nice that it is self contained too.
Your code was all I needed to solve problems in my codes due to too many useless points. Thanks a lot!
Andrii, just out of curiosity, what would you want to get from: https://maps.google.co.uk/maps?q=guggenheim+museum+bilbao&hl=en&ll=43.268728,-2.933803&spn=0.001457,0.002902&sll=43.267445,-2.937035&sspn=0.002914,0.005804&t=h&hq=guggenheim+museum+bilbao&z=19
Thank you for the link.
Not sure about Matlab implementation, but you can find C implementation if you follow the link provided on this page:
It is for building extraction from aerial images. After segmentation process boundaries are zigzag-shaped with many intrusions. I would like to simplify polygons but to maintain their essential shape. It means buildings should have orthogonal angles and small extrusions should be cut off. So I am looking for some implementation algorithm in MatLab. Like this: http://img812.imageshack.us/img812/3284/exampleus.png
@ Andrii, what is it you want to achieve by orthogonalizing a set of input points? Could you clarify what this operation means or perhaps provide a reference?
Hi Anton. I have tried your function.
It works very well.
Do you know about some function like your but
which can simplify and after orthogonalize input set of points?
- migrated to GitHub
Modified code according to suggestions made by Georgios Gkantzounis and Erik Husby