joakimlindblad / imhausdorff Goto Github PK

This is nice code and useful, but I'm not sure it will give values that correspond to what you would get when explicitly looking at point correspondence on the boundaries of two regions. In particularfor binary label maps A & B, as an image and its negation share the same boundary, imhausdorff(A,B) should be equal to imhausdorff(~A,~B). This is not guaranteed with the current code.

For instance the following script constructs two circular images centred on the same point with different radii and some modifications. The larger circle has a short additional linear segment added to the outside and the smaller circle has a longer linear segment removed. In this toy example imhausdorff only picks up on one of the lines resulting in two different returned distances for the images and their negations.

Atemp = false(127);
Atemp(64,64)=true;
Btemp = Atemp;

se = strel('disk',15,0);
Atemp = imdilate(Btemp, se);

se = strel('disk',20,0);
Btemp = imdilate(Btemp, se);

Atemp(63:65,64:end) = false;
Atemp(63:65,64) = true;
Btemp(63:65,40:64) = true;

[D1,indx] = imhausdorff(Atemp,Btemp);

[I11,J11]=ind2sub([127,127],indx(1));
[I12,J12]=ind2sub([127,127],indx(2));

[D2,indx2] = imhausdorff(~Atemp,~Btemp);
[I21,J21]=ind2sub([127,127],indx2(1));
[I22,J22]=ind2sub([127,127],indx2(2));

Hello,
First, thank you for this useful work. What is the unit of the result got as output as a result of the code? mm or pixel?
Best regards,