Non-negative Matrix Factorization (NMF) is one of the recently emerged dimensionality reduction methods. Unlike other methods, NMF is based on non-negative constraints. These constraints lead to a parts-based representation because they allow only additive, not subtractive combinations. The non-negative basis vectors can represent the local features of the data.Facial expression recognition is one of the most challenging problems in the fields of pattern recognition, machine vision, affective computing and psychology. It has turned into an active research topic in the recent decade. In this paper, Non-negative Matrix Factorization has been applied to the facial expression database for extracting the features of the facial expressions. A Principal Component Analysis (PCA) approach was performed as well for facial expressions recognition for comparison purposes. Because the outputs of NMF are localized features, we can use these parts based features collectively to represent a face. With the underlying non-negative constraints, basis images and corresponding encoding are both non-negative. In NMF, as the name implies, the non-negative adds constraints to the matrix factorization, allowing only additions in the synthesis, there are no cancellations or interference of patterns via subtraction or negative feature vector values. Then we use nearest neighbor to classify the facial expressions by the encoding of the base images.We found that, for the CMU database, NMF outperforms PCA, and the experiment on the facial databases testify the validity of this method. |