A Label-Embedding Online Nonnegative Matrix Factorization Algorithm
Zhibo Guo, Ying Zhang · IEEE Access · 2019
Nonnegative matrix factorization is a widely used data processing method, which has been applied in many fields, such as data dimension reduction and feature extraction. Considering the label information of training samples is helpful to improve the performance of data dimension reduction and classification, and how to apply the subspace calculated by training samples to testing samples to improve the computation efficiency is also a problem worth considering. In this paper, an algorithm called label-embedding online nonnegative matrix factorization (LEONMF) is proposed for image dimensionality reduction and classification. First, the label is embedded in the training matrix, by which the base matrix and the weight matrix can be calculated. The data amount of the base matrix is greatly small than that of the input matrix data. Next, the base matrix is adjusted and connected with the testing matrix to get the new connection matrix. The new connection matrix can be expressed as the product of the new base matrix and the new weight matrix. The new weight matrix can be used for classification after adjustment. The experimental results demonstrated that the proposed LEONMF outperforms the state-of-the-art algorithms in classification accuracy and calculation speed.