Assumption: There's always one and only one non-zero element in each matrix, and the matrices have the same size
assert(all(sum(ob ~= 0, 2) == 1) && all(sum(tb ~= 0, 2) == 1)) %check 1st assumption
assert(isequal(size(ob), size(tb))) %check 2nd assumption
Step 1: find the indices of non-zero column in each row using find. Because find works column-wise, transpose the matrices and the problem becomes: find non-zero row in each column:
[obcol, ~] = find(ob'); %the ~ is required as you want the two return value version of find
[tbcol, ~] = find(tb');
Step 2: compare the indices and when they differ extract the corresponding element from each matrix using sub2ind:
differcol = obcol ~= tbcol;
rowindices = (1:size(ob, 1))';
rowindices = rowindices(differcol); %indices of row where non-zero column differ
obvalues = ob(sub2ind(size(ob), rowindices, obcol(differcol)));
tbvalues = tb(sub2ind(size(tb), rowindices, tbcol(differcol)));
Step 3: subtract the differences and sum it all:
diffsum = sum(obvalues - tbvalues)
