Vectoring the blurring of an image, with use of a kernel

Hello,

I have to blur an image with use of a kernel. Currently I have several for loops, however, this is slow. I was wondering how I could go about vectorising it while only using functions from base matlab (no toolboxes unfortunatly). I'm very new to matlab so any help would be greatly appreciated. My current code is below:

inputImage = imread("testImage2.jpg");
inputImage = double(inputImage);
w = 5;              % specifies the size of the box (w = 3 means a 3x3 box)
r = fix(w/2);       % 'radius' of box, excluding centre pixel

[x,y,z] = size(inputImage);             % 3D sizes of image
blurredImage = zeros(size(inputImage)); % initialize output array

for i=(r+1):x-(r+1)             % scan through every pixel in inputImage
    for j=(r+1):y-(r+1)         % '(rad+1)' leaves a border so data isn't read outside array bounds
        for (k=1:3)             % scan through rgb values of each pixel
            kernel = inputImage([i-r:i+r],[j-r:j+r], k);     % array index the pixel data within the box around the current pixel
            blurredImage(i,j,k) = mean(kernel(:));           % apply mean of pixel data within box to blurredImage pixel
        end
    end
end
figure(1)
image(uint8(inputImage))
figure(2)
image(uint8(blurredImage))

[Zachary Cockshutt]

I had an attempt at vectorising and came up with this, however, it just ouputs a grey image:

i1 = double(imread("IMG_5704.JPG"));
dim = size(i1);
iOutput = zeros(size(i1));

w = 5;              % specifies the size of the box (w = 3 means a 3x3 box)
r = fix(w/2);       % 'radius' of box, excluding centre pixel

i = (r+1):dim(1)-(r+1);
j = (r+1):dim(2)-(r+1);
kernel = i1([i-r:i+r],[j-r:j+r],:);
iBlur(i,j,:) = mean(kernel,'all');

Answer 1

[Raunak Gupta]

Hi,

Without Toolboxes such as Image Processing Toolbox I see the only way achieving the kernel multiplication with the help of conv2. Here since the kernel is doing averaging that is why convolution definition will not alter the result. Using the ‘shape’ name-value as ‘same’ you may get the blurred image. Use the kernel as mentioned below based on window size. Apply the same kernel on each channel and final image should be same as done by imfilter.

windowSize = 5; 
kernel = ones(windowSize) / windowSize^2; 

[Zachary Cockshutt]

Yep this is perfect, cheers for the help mate