Fig. 4From: Improving clustering performance using independent component analysis and unsupervised feature learningVisualization of the USPS clusters using multidimensional scaling. (MDS). MDS is performed on the similarity matrix of the observations, which is also used in GNMF clustering. a Clusters based on RICA extracted features are more distinct and homogeneous. b Clustering on the pixels only after applying L2-normalization did not provide sufficient separation of the clustersBack to article page