Unsupervised Discrimination of Clustered Data via Optimization of Binary Information Gain