Non Negative Matrix Factorization Clustering Capabilities; Application on Multivariate Image Segmentation

• Main Authors: Cosmin Lazar, Danielle Nuzillard, Patrice Billaudel, Sorin Curila
• Format: Article
• Language: English
• Published: Editura Universităţii din Oradea 2009-10-01
• Series: Journal of Electrical and Electronics Engineering
• Subjects: Classification; metrics; non negative matrix factorization; segmentation; multivariate
• Online Access: http://electroinf.uoradea.ro/reviste%20CSCS/documente/JEEE_2009/Articole_pdf_JEEE_EL_nr_2/JEEE_2009_Nr_2_EL_Lazar_NonNegative.pdf

The clustering capabilities of the Non Negative MatrixFactorization algorithm is studied. The basis images are consideredlike the data membership degree to a particular class.A hard clustering algorithm is easily derived based on theseimages. This algorithm is applied on a multivariate image toperform image segmentation. The results are compared withthose obtained by Fuzzy K-means algorithm and better clusteringperformances are found for NMF based clustering. We also showthat NMF performs well when we deal with uncorrelated clustersbut it cannot distinguish correlated clusters. This is an importantdrawback when we try to use NMF to perform data clustering.