Neonatal Pain Expression Recognition Based On Block Weighted Local Binary Patterns

Currently, the assessment of neonatal pain is assessed by specially trained medical staff, this method takes a lot of time and energy, and the assessment results are often affected by subjective factors. Therefore, the development of a automatic neonatal pain facial expression assessment system is very meaningful.For pain expression recognition of newborns, this paper presents neonatal pain expression recognition based on block weighted local binary patterns. The main research work and achievements are as follows:(1) The influence of different block sizes of recognition performance when using uniform pattern and symmetric pattern LBP feature extraction. Against neonatal pain expression image doing LBP conversion, using different block sizes, using uniform and symmetrical patterns LBP feature extraction.(2) Through observing a large number of various types of neonatal pain emoticons, analysis of the contribution of different facial expressions organs or regions for the classification, to highlight its expression characteristics, giving greater weight to higher level dependence areas. Use regular and irregular block pattern for feature extraction.(3) Research on the characteristics of PCA,2DPCA and2DLDA dimension reduction methods. In the experiment of this paper, use PCA method for dimension reduction. Analyze the relationship between PCA feature dimension and recognition rate.(4) Research on the classification method based on sparse representation of neonatal pain expression.The results showed that:The LBP-based feature extraction block recognition rate has improved significantly; Comparing with non-weighted recognition rate, the weighted block recognition rate than has improved; When it have600training sample images, feature extraction using block based weighted LBP obtain the highest average recognition rate (84.75%).