Umeå University's logo

Facial expressions are considered as a reliable indicator in neonatal pain assessment. This paper proposes a novel recognition method for neonatal pain expression. The method can utilize the feature descriptors based on the weighted local binary pattern (LBP) and the classifier based on sparse representation. Firstly, the normalized facial image is described using a feature vector, which is histogram sequence obtained by concatenating the weighted histograms of the LBP feature maps of all the local blocks. Then, the principalc component analysis (PCA) method is used to reduce the dimensions of the feature vector of training and test samples. Finally, the over-complete dictionary is built and the classifier based on sparse representation is used to classify test samples into four classes of facial expressions: calm, crying, mild pain, and severe pain. The objective of this study is to assist the clinicians in assessing neonatal pain by utilizing computer-based image analysis techniques. Experimental results on neonate facial image database show the effectiveness of the proposed method. The classification accuracy rate reaches 84.50%.