[learning notes] matlab self compiled image convolution function

Image convolution principle

Code

% Convolution function 
%made by yao
function result = myconv(kernel,img)
    [k,num] = size(kernel);
    % Determine whether the passed in array is double precision floating-point , Otherwise, type conversion 
    if ~isa(img,'double')
        p1 = double(img);
    else
        p1 = img;
    end
    % Extract dimensions 
    [m,n] = size(p1);
    % Construct the template after convolution 
    % Here we consider the problem of edges , So the black edge was cut off 
    result = zeros(m-k+1,n-k+1);
    for i = ((k-1)/2+1):(m-(k-1)/2)
        for j = ((k-1)/2+1):(n-(k-1)/2)
            result(i-(k-1)/2,j-(k-1)/2) = sum(sum(kernel .* p1(i-(k-1)/2:i+(k-1)/2,j-(k-1)/2:j+(k-1)/2)));
        end
    end
end

Here's an example , For example, we have 3x3 Convolution kernel , Convolute the image , So in （1,1） When , It will be associated with the outside of the image edge （ Including the upper row and the left column , Similarly, the same situation will appear at the bottom and right of the image ）.
So to avoid this , Will the 3x3 The starting point of convolution is set at (2,2) It's about . If the image has m Column n That's ok , Then the convolution of one line terminates in n-1 Out , The convolution end point of a column is m-1 It's about .
Empathy ,5x5 And longer convolution kernels are the same operation .
5x5 The convolution kernel of starts at （3,3）, Termination point （m-2,n-2）.
So for odd convolution kernels , Let the length of convolution kernel be k, Then the initial convolution point is ((k-1)/2+1), Termination point （m-(k-1)/2）（ The ranks are the same , Just change it ）.