BP Neural Network-Based Deep Non-negative Matrix Factorization for Image Clustering

Deep non-negative matrix factorization (DNMF) is a promising method for non-negativity multi-layer feature extraction. Most of DNMF algorithms are repeatedly to run single-layer NMF to build the hierarchical structure. They have to eliminate the accumulated error via fine-tuning strategy, which is, however, too time-consuming. To deal with the drawbacks of existing DNMF algorithms, this paper proposes a novel deep auto-encoder using back-propagation neural network (BPNN). It can automatically yield a deep non-negative matrix factorization, called BPNN based DNMF (BP-DNMF). The proposed BP-DNMF algorithm is empirically shown to be convergent. Compared with some state of the art DNMF algorithms, experimental results demonstrate that our approach achieves superior clustering performance and has high computing efficiency as well.