Matlab：从代码中删除循环

Question

flow 是一个 4D 双矩阵，其第四维总是2。我想在不使用循环的情况下实现这个循环。

本质上，对于第三维中的每个索引，我想将flow1 划分为大小为block-size 的正方形块，然后计算每个块的第四维值的平均值，然后将这些值分配回该块中所有元素的相应第四维。

典型值为 [height width N block_size] = [ 480 640 100 8] 。保证block_size 是height 和width 的一个因子

[height,width,N,~]  = size(flow1);
reduced_flow = zeros(size(flow1));
for ii = 1:block_size:height
    for jj = 1:block_size:width
        for k = 1:N
            reduced_flow(ii:ii+block_size-1,jj:jj+block_size-1,k,1) = mean(mean(flow1(ii:ii+block_size-1,jj:jj+block_size-1,k,1),1),2);
            reduced_flow(ii:ii+block_size-1,jj:jj+block_size-1,k,2) = mean(mean(flow1(ii:ii+block_size-1,jj:jj+block_size-1,k,2),1),2);
        end
    end
end

示例

 flow1 = round(10*rand(4,4,2,2));
 block_size = 2;
 [flow1 reduced_flow]

 ans(:,:,1,1) =

     6.0000    4.0000    1.0000         0    4.0000    4.0000    2.2500    2.2500
     6.0000         0    2.0000    6.0000    4.0000    4.0000    2.2500    2.2500
          0    5.0000    1.0000    3.0000    3.2500    3.2500    2.7500    2.7500
     6.0000    2.0000    2.0000    5.0000    3.2500    3.2500    2.7500    2.7500


 ans(:,:,2,1) =

     7.0000    1.0000    3.0000    4.0000    4.5000    4.5000    2.7500    2.7500
     5.0000    5.0000    2.0000    2.0000    4.5000    4.5000    2.7500    2.7500
     5.0000    9.0000    6.0000    9.0000    6.7500    6.7500    5.5000    5.5000
     4.0000    9.0000    6.0000    1.0000    6.7500    6.7500    5.5000    5.5000


 ans(:,:,1,2) =

     1.0000    9.0000    7.0000    5.0000    3.2500    3.2500    7.5000    7.5000
     2.0000    1.0000   10.0000    8.0000    3.2500    3.2500    7.5000    7.5000
     7.0000   10.0000    3.0000    8.0000    6.2500    6.2500    4.0000    4.0000
     3.0000    5.0000    4.0000    1.0000    6.2500    6.2500    4.0000    4.0000


 ans(:,:,2,2) =

     2.0000    3.0000    7.0000    7.0000    2.7500    2.7500    6.5000    6.5000
     4.0000    2.0000    5.0000    7.0000    2.7500    2.7500    6.5000    6.5000
     1.0000    2.0000    9.0000    6.0000    4.2500    4.2500    4.5000    4.5000
     5.0000    9.0000    1.0000    2.0000    4.2500    4.2500    4.5000    4.5000

Answer 1

方法 1

flow11 = reshape(flow1,size(flow1,1),[]);
fun = @(block_struct) mean2(block_struct.data);
mat1 = blockproc(flow11,[block_size block_size],fun);
mat2 = imresize(mat1,block_size ,'nearest');
reduced_flow = reshape(mat2,size(flow1));

方法二

t1 = sum(reshape(sum(reshape(flow1,block_size,[])),height/block_size,block_size,[]),2)./block_size^2;
t2 = reshape(t1,1,1,[]);
t3 = t2(ones(1,block_size),ones(1,block_size),:,:);
t4 = reshape(permute(t3,[1 3 2]),size(t3,1)*size(t3,3),[]);
out = permute(reshape(t4,height,size(t4,1)/height,[]),[1 3 2]);
reduced_flow = reshape(out,size(flow1));