How to replace for-loop with matrix multiplication?

I'm trying to replace for-loop with matrix multiplication in order for efficiency. In my case, the matrix is 3-dimensional. I'm not sure how to achieve it. I've attached my code. Could someone help optimize my code?

[Bruno Luong]

why reinvent the wheel and not using MATLAB interpolation functions?

[Duen Chian]

This is an assignment to practice the implementation of bilinear interpolation, so I cannot use any relevant function like imresize().

Answer 1

[Nipun]

Hi Duen,

I understand that you are trying to replace the for loops in the given code with matrix multiplication for imroved time efficiency. I assume that the matrix that you are referring to in the question is a different 3 dimensional matrix, since resizedImage is a 2 dimensional matrix.

Here are few steps that I followed to optimize the code:

1. vectorize variables a and b
2. generate the weights as a vector
3. Use element-wise operations for the matrix elements and reshape the matrix to get the desired image size

Please note that the optimized code below assumes that the img matrix contains grayscale pixel values. If you're working with RGB images, you'll need to modify the code accordingly.

function resizedImage = resizeImage_bilinear_Opt(originalImage, scalingFactor)
    % load image
    img = imread(originalImage);

    % get source row and column
    [sr, sc] = size(img);

    % get destination row and column
    dr = ceil(sr * scalingFactor);
    dc = ceil(sc * scalingFactor);

    % generate row and column indices for the destination image
    a = (1:dr) * ((sr - 1) / (dr - 1));
    b = (1:dc) * ((sc - 1) / (dc - 1));

    % compute the floor and ceil values for row and column indices
    x = floor(a);
    y = floor(b);
    x1 = min(x + 1, sr);
    y1 = min(y + 1, sc);

    % compute the weights for bilinear interpolation
    wa = a - x;
    wb = b - y;
    wa1 = 1 - wa;
    wb1 = 1 - wb;

    % perform bilinear interpolation using matrix multiplication
    resizedImage = uint8(wa1 .* wb1 .* double(img(x, y)) + ...
                        wa1 .* wb .* double(img(x, y1)) + ...
                        wa .* wb1 .* double(img(x1, y)) + ...
                        wa .* wb .* double(img(x1, y1)));

    % reshape the image to the desired size
    resizedImage = reshape(resizedImage, dr, dc);
end

Hope this helps.

Regards,

Nipun