How can I vectorize/LUT this for loop ?

Hello,

I have been dabbling trying to make a look-up table mapping for this tedious for-loop but I wasn't successful. Here is the code and explanation of what I am trying to do:

    for kk=1:size(aa,1)
    imageModified(aa(kk,1),aa(kk,2))=img(aa(kk,3),aa(kk,4));
    end

My 'kk' is in order of millions for this loop is somehow time consuming.

Basically, I have a matrix called 'aa'.

'aa' is some Nx4 matrix where each row has mapping values.

What I want to do is going through row by row of matrix 'aa' and map image value 'img' to its corresponding place in image 'imageModified'.

For example,

Let's say aa=[ 50 40 60 70 ; 2 56 28 10 ; .... ]

So what I want to do is: map img(50,40) value to location imgModified(60,70). Map map img(2,56) value to location imgModified(28,10). And do that for 'aa' N rows.

Thank you in advance,

Answer 1

[Sven]

Hi Louis,

This one's got a nice solution using sub2ind that lets you do the whole thing at once without a loop:

img = rand(100,100)
imgModified = zeros(size(img))
aa = [ 50 40 60 70 ; 2 56 28 10]; % etc
imgInds = sub2ind(size(img),aa(:,1),aa(:,2))
modInds = sub2ind(size(img),aa(:,3),aa(:,4))
imgModified(modInds) = img(imgInds)

Does that end up with what you intended?

[Louis]

Awesome. Thank you very much, Sven. Time down from 22 seconds to 0.2 seconds :)

[Sven]

No problems, you get a vote for writing a nice clear question with example code :)