hi vasilis,
have a look through this: rather than using a loop (which should work fine, but probably slower) you can use indexing. comments below hopefully helpful; since you describe matrices of different sizes, there's an extra step to make them equal. there's lots of equivalent ways of working around this, depending how many assumptions you want to make (ie are the data always the same size, etc).
%% this part shows the logic
% simple 1D or 2D data (equal size)
FinalProduct = NaN(1,6); FinalProduct(4) = 45; B = NaN(1,6); B(3) = 12;
% FinalProduct = NaN(5); FinalProduct(find(eye(5))) = 123; B = fliplr(FinalProduct)/2;
% these are the indices you want to replace
targetIndices = find(isnan(FinalProduct) .* ~isnan(B));
%replace these indices in a using values from b
NewResult = FinalProduct; % just to illustrate
NewResult(targetIndices) = B(targetIndices);
% display output for sanity check
FinalProduct
B
NewResult
%% for the data you described:
% for 3D data of unequal dimensions, 'easiest' (?) to make dummy for B
% this allows the earlier logic to work exactly the same
% strictly, this assumes FinalProduct > B, otherwise need to 'fill out' smaller to fit larger
%% Read in your data here
% FinalProduct =
% B =
sizeOfFinalProd = size(FinalProduct);
sizeOfB = size(B);
dummyB = NaN(sizeOfFinalProd);
dummyB(1:sizeOfB(1),1:sizeOfB(2),1:sizeOfB(3)) = B; % or as many dimensions as needed :)
% now FinalProduct and dummyB are the same size, with B padded out with NaN
targetIndices = find(isnan(FinalProduct) .* ~isnan(dummyB));
NewResult = FinalProduct;
NewResult(targetIndices) = dummyB(targetIndices);
