How to create 2D sub arrays by sampling every n by n points

Hi, I want to create 2D sub arrays by sampling every n by n points. Specifically, I have an array with size of 280x280. I want to sample every 8 by 8 points and form a 3D array with size of (35,35,64).

Currently I am using loop (refer to below) which takes abut 0.25msec on my Laptop. I am wondering if there is a more elegent and faster way to create the 3D array by taking advantage of matlab internal indexing capability.

Thank you!

%%%%%
%A=rand(280,280);
A = double(imread('AT3_1m4_01.tif'));
A = A(1:1:280,1:1:280)/256;
B=zeros(35,35,64);

numIter=1000;
tic
for iter=1:numIter 
    for i=1:1:8
        for j=1:1:8
            indx1=(i-1)*8+j;
            B(:,:,indx1)=A(i:8:end,j:8:end);
        end
    end
end
disp(toc/numIter);

figure(1); montage(B);
%%%%%%%%%%%%%%%%%%%%%%%%%%

Answer 1

[Yiwu Ding]

Thank you all for your comments and time.

Referring to the following updated code, was able to improve the code speed from 0.25msec to 0.14msec by using the vector indexing.

Note iter of 1000 is used to get more accurate timing for single run.

Plots are used for data verification.

close all;
clear;

%A=rand(280,280);
A = double(imread('AT3_1m4_01.tif'));
A = A(1:1:280,1:1:280)/256;
B=zeros(35,35,64);

%%%%Approach#1 iteration
numIter=1000;
tic
for iter=1:numIter 
    for i=1:1:8
        for j=1:1:8
            indx1=(i-1)*8+j;
            B(:,:,indx1)=A(i:8:end,j:8:end);
        end
    end
end
toc
disp('Approach#1 time: single iter')
disp(toc/numIter);

figure(1); montage(B);


%%%%%%%%%%%%%%%%%%%%%
%%%%Approach#2 vector indexing
tic
for iter=1:numIter 
    rngY=1:280;
    rngY_permute=reshape(rngY,8,35);
    rngY_permute=rngY_permute';
    B1=A(rngY_permute,rngY_permute);
    B2=reshape(B1,35,8,35,8);
    B3=permute(B2,[1,3,4,2]);
    B4=reshape(B3,35,35,64);
end
toc
disp('Approach#2 time: single iter')
disp(toc/numIter);

figure(2); montage(B4);

figure(3); montage(B4-B); title('\Delta');

max(max(max(abs(B4-B))))

[Adam Danz]

You're adding the iter-loop just to add extra time to your computation for whatever reason. There are much better ways to building in time if you need to add delays.

You could use a timer function or the pause() command. Both offer much better control over timing and much less computational cost than this "Approach #2" .

This is not a good solution to your problem.

Answer 2

[Ridwan Alam]

B = reshape(A,35,35,[]);

[Adam Danz]

The indexing isn't consecutive, though.

B(:,:,indx1)=A(i:8:end,j:8:end);

[Adam Danz]

B = permute(A,dimorder) rearranges the dimensions of A. The loop in the OP's code is extracting every 8 rows and every 8 columns so it's not clear to me how permute() would help.

Answer 3

[Adam Danz]

The iter-loop is completely unnecessary and is merely doing the same exact work 1000 times.

numIter=1000;
tic
% for iter=1:numIter         Remove this line
    for i=1:1:8
        for j=1:1:8
            indx1=(i-1)*8+j;
            B(:,:,indx1)=A(i:8:end,j:8:end);
        end
    end
% end                        Remove this line

Removing that reduces the processing time to 0.000047 sec. Since you're indexing non-consecutively, the loop method (without the iter-loop) will be quite fast and I doubt an ugly vectorized approach will be faster.