Help with average of elements in a matrix

I have a matrix M with dimension 630x500. I want to do an average of its values along vertical direction, in order to have a matrix with smaller dimension. For example, I want to obtain a matrix with dimension 210x500: this means that, every three values along each column, I take just one (which is the average of the three values). I do this with the following code:

n=3
M = squeeze(mean(reshape(M,n,[],size(M,2))));

Let's suppose that, instead of three, I want to do an average over four values, i.e. I take just one value which is the average of the four values. This line

n=4
M = squeeze(mean(reshape(M,n,[],size(M,2))));

doesn't work, because I cannot divide 630 by four ! How can I find a way to approximate best this requirement ? For example, I can do an average over the first 628 elements (628 contains 4) and let the last two elements unchanged. How can I do this automatically once I have chosen an integer n ?

Answer 1

[Andrei Bobrov]

n = 3;
s = size(M);
out = squeeze(nanmean(reshape([M;nan(mod(-s(1),n),s(2))],n,[],s(2))));

without nanmean :

s = size(M);
n = 3;
k = rem(s(1),n);
M1 = [M;zeros(n-k,s(2))];
out = squeeze(bsxfun(@rdivide,...
            sum(reshape(M1,n,[],s(2))),[repmat(n,1,floor(s(1)/n)),k]));

other variant

M2 = conv2(M,[1;1;1]/3);
out = M2(n:n:end,:);

Answer 2

[Iain]

 n=4
 elements = numel(M);
 sets = round( elements / n / size(M,2));
 elements_to_use = sets * n * size(M,2);
 Mnew = squeeze(mean(reshape(M(1:round(numel(M(1:elements))/(n*size(M,2)) ) ,n,[],size(M,2))));
 Mnew(:,(end+1):(end+elements - elements_to_use)) = M((elements_to_use+1):end);

[aurc89]

It gives me this error:

Index exceeds matrix dimensions.

for the line

Mnew = squeeze(mean(reshape(M(1:round(numel(M(1:elements))/(n*size(M,2)) ) ,n,[],size(M,2))));

whatever is n

[Iain]

Whoops:

 Mnew = squeeze(mean(reshape(M(1:elements_to_use),n,[],size(M,2))));

Answer 3

[Image Analyst]

If you have the Image Processing Toolbox, simply do

resizedMatrix = imresize(M, [230, 500]);

There are a variety of averaging and interpolation techniques you can choose from as third input arguments.