Sparse symmetric nonnegative matrix factorization applied to face recognition

The task of Sparse Symmetric Nonnegative Matrix Factorization(SSNMF) is formulated as optimization problem and solved numerically with the method of projected gradients descent. The adjustable sparsity level allows to emphasize the most significant object features. Clustering of the Yale Faces data set shows that SSNMF provides the same level of quality as common clustering approaches.