juanji2_2d.m

clc; clear;

x = rand(7,7);     % 卷积核——必须是方阵且为奇数行/列
data = rand(8,7);  % 数据——最好比卷积核的尺寸大

zidai = conv2(data,x,'same')

x = rot90(rot90(x));  % 新的卷积核

% 核的尺寸
size_x = size(x);
row_x = size_x(1);  % 核的行数
col_x = size_x(2);  % 核的列数
% 数据的尺寸
size_data = size(data);
row_data = size_data(1);  % 数据的行数
col_data = size_data(2);  % 数据的列数

% 核的中心元素:
centerx_row = round(row_x/2);
centerx_col = round(col_x/2);
centerx = x(centerx_row,centerx_col);

% 对原始数据扩边:
data_tmp = zeros(row_data+row_x-1,col_data+row_x-1);
data_tmp(centerx_row:centerx_row+row_data-1,centerx_row:centerx_row+col_data-1) = data;
data_k = data_tmp;
% 扩边后新数据矩阵尺寸:
size_data_k = size(data_k);
row_data_k = size_data_k(1);
col_data_k = size_data_k(2);

% m = centerx_row:row_data+row_x-2
% 开始卷积计算: m n 是新数据矩阵的正常索引
result = zeros(row_data_k,col_data_k);

for m = centerx_row:centerx_row+row_data-1
    for n = centerx_row:centerx_row+col_data-1
        % tt是临时与卷积核大小相同的数据中的部分矩阵:
        tt = data_k(m-(centerx_row-1):m+(centerx_row-1),n-(centerx_row-1):n+(centerx_row-1));
        % juan是中间每一次卷积计算求和的中间量:
        juan = sum(x.*tt);
        result(m,n) = sum(juan(:));
    end
end

% 求掉之前扩边的0:
result = result(centerx_row:centerx_row+row_data-1,centerx_row:centerx_row+col_data-1)