Nonnegative matrix factorization (NMF) is a recent method used to decompose a given data matrix into two nonnegative sparse factors. There are many techniques applied to enhance abilities of NMF, particularly kernel technique which discovering higher-order correlation between data points and obtaining more powerful latent features. This paper presents an overview of kernel methods on NMF along with its representation and recent variants. The development as well as algorithms for kernel based NMF are discussed and presented systematically.