Question

I have a Matlab Code Snippet and I try to write in C++, but I really don’t understand
what is happening there:

for c = 1:3
    Id = double(I(:,:,c))/255;
    Wc(:,:,c) = sum(Id(pixels).*weights, 3);
end

There is an image I of size 480x640x3. In the first Iteration the first channel of the image is
saved in Id, which then has a size of 480×640. But what is happening in the next line? I just don’t
understand that syntax.

pixels and weights are of size 300x383x4x1.

So what does this line exactly do?:

sum(Id(pixels).*weights, 3);

Thank you.

Answer 1

sum(X, n) summarizes all elements of a matrix X across the n-th dimension, so this:

sum(Id(pixels) .* weights, 3);

really does the following:

• Extracts a sub-matrix from matrix Id, pixels being the indices of the extracted values:
  matrix Id is imlicitly converted to a column vector, and the resulting sub-matrix is of the same size as pixels (each value being equal to Id(p), where p is the corresponding element in pixels).
• Multiplies the extracted matrix by another matrix weights element-by-element (note that the multiplication operator is .*)
• Summarizes all elements of the resulting 3-D matrix along the third dimension.

You can look into the official documentation for more information.

EDIT: A Simple Example

Let’s assume that:

Id = [10 20; 30 40; 50 60; 70 80];
pixels(:, :, 1) = [4 4; 4 4];
pixels(:, :, 2) = [6 6; 6 6];
pixels(:, :, 3) = [8 8; 8 8];

that is, Id is a 2-D matrix, and pixels is 3-D. Now, Id(pixels) would yield:

ans(:, :, 1) =

  70  70
  70  70

ans(:, :, 2) =

  40  40
  40  40

ans(:, :, 3) =

  80  80
  80  80

because if you convert Id to a column vector (try Id(:)), 70 is the 4^th element, 40 is the 6^th, and 80 is the 8^th. Note that the result has the same dimensions as pixels (not Id!).

Hope that helps!