CODEWITHSUNDEEP
matlabmultidimensional-arraymeanpca
cathy305
cathy305

Reputation: 57

Subtract mean from multi-dimensional data

I want to do a Principal Component Analysis for a large number of samples. I have no problem subtracting the mean from audio samples, since audio only has 2 dimensions and I can easily use a for loop .

However, this is a different case for video since each video sample has about 18-20 dimensions.

Example of content of one video file: whos -file sample_video_001.mat result: size: 54x96x19. bytes: 98496 . class: uint8. attributes: -

How can I compute this?

Upvotes: 0

Views: 397

Answers (1)

Adriaan
Adriaan

Reputation: 18177

You can use the mighty bsxfun to calculate the mean per dimension and directly subtract it from the original array.

A = randi(256,54,96,19,'uint8');   %// Some random data, replace with your own
B = double(A);                     %// Cast data to double
Bav = bsxfun(@minus,B,mean(B,3));  %// Subtract the mean

It turned out to be a bit more complicated that I initially thought, as you have a 'uint8' class matrix. The mean of your data along the third dimension will not be an integer most likely, and will therefore be automatically set to class 'double', failing a direct bsxfun. If you first convert your original data to 'double' and then use bsxfun it will work. Possibly you might have to divide by 256 to get data in the range [0 1] to allow MATLAB to recognise it as a plottable format (so do B = double(A)./256;). You cannot go back to 'uint8', since you subtract a non-integer mean from your data, so the result will not be integer either.

There's a function called pca as well though, which is probably more suited to what you need, as it is a build-in function. Be sure that you know how to use it properly.

Upvotes: 4

Related Questions