On Classification of Bivariate Distributions Based on Mutual Information