A method is presented for similarity search based on dimensionality reduction by Non-negative Matrix Factorization (NMF). In similarity search based on NMF, the retrieval accuracy depends on the parameters e.g. the initial values of the factor matrices, the matrix rank and the number of iterations. In our method, we set the absolute value of elements of the orthonormal matrices obtained by Singular Value Decomposition (SVD) to the initial values of the matrices in NMF. This initialization method helps us to determine the parameters in NMF. It is verified by experiments of image and video search that the retrieval accuracy by the proposed method is better than that by Latent Semantic Analysis (LSA).
展开▼