Efficient Algorithms for Regularized Nonnegative Scale-invariant Low-rank Approximation Models