Probably as simple a solution as any is to return the indices where the change occurs with both row and column in a form that doesn't depend on the number...unless, of course, there are none globally.
[r,c]=find(diff(A)==1);
NB: You may want to do a "+1" to the above for the row since the difference vector is one element shorter than the original.
You then iterate over the values of c to find the rows for each column for which there is at least one transition as
rows=arrayfun(@(i)r(c==i),unique(c),'UniformOutput',false);
Of course, you can look for any one column as well by
rows=r(c==2);
will be the second column
