Can you use islocalmax2 for a nxnxn grid where the values of that grid correspond to peaks in data density?

I am trying to pickout peaks of data density. I fit my data to a kernel density estimation using the mvksdensity function.

I then get a matrix, Z, the size of my grid where the values correspond to the density of the data in the 3D variable space.

My question is, can I use islocalmax2 to pick out peaks in this nxnxn matrix Z, or does it only work in 2 dimensional arrays.

The documentation lists the input data can be a multidimensional array, though I'm not sure if it works with a page size greater than 1.

My goal is to calculate the prominence of of these peaks as well so if anyone has any resources related to that in more than 2 dimensions, that would be very helpful.

I've also used the imregional max function from the image processing toolbox, but the user doesn't have control over any parameters for the peaks.

Answer 1

[Milan Bansal]

Hi Poison Idea fan,

The function islocalmax2 is compatible with multidimensional arrays, that means you can use it to get the peaks in the n x n x n matrix "Z" as well. The function will return the output as a 3 dimensional "logical" array with same size as that of "Z". The function will find local maximas for each 2-D layer (n x n) in "Z".

A = rand(100, 100, 100); % create a random 3D matrix of size 100 x 100 x 100
TF = islocalmax2(A); 
size(TF)

ans = 1x3

100 100 100

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

You can also calculate the "prominence" using the same function as shown below.

[TF, P] = islocalmax2(A); % P is the prominence

Please refer to the following documentation link to learn more about islocalmax2 function.

https://www.mathworks.com/help/matlab/ref/islocalmax2.html

Hope this helps!

[Poison Idea fan]

Hi Milan,

The issue for me is that it's finding the LM for each layer and not considering the relative values in the Z direction. Unless you have a way to suggest taking the maxima of each 2D layer and comparing them somehow, I am still stuck.

PIF