摘要： 为提高疼痛表情的识别准确性，将约束局部神经域（CLNF）模型和贝叶斯网络（BN）模型相结合，提出一种疼痛表情识别方法。利用CLNF模型获取疼痛表情的关键特征点，在此基础上得到携带大量疼痛信息的面部活动单元（AU），通过对AU加标签处理得到样本数据集。根据定性专家经验获取BN条件概率之间的约束集合，采用变权重方法将样本数据集与约束扩展参数集相融合以完成BN模型的参数估算，并通过BN推理方法实现疼痛表情的最终识别。实验结果表明，与概率潜在语义分析、局部二值体卷积神经网络等方法相比，该方法可有效提高疼痛表情的识别性能，具有更高的识别精度。

关键词: 疼痛表情识别, 约束局部神经域模型, 活动单元, 贝叶斯网络模型, 变权重融合, 参数估算

Abstract: In order to improve the accuracy of pain expression recognition,this paper proposes a pain expression recognition method that combines the Constrained Local Neural Field(CLNF) model with the Bayesian Network(BN) model.The CLNF model is used to obtain the key feature points of pain expressions and on this basis acquire the facial Activity Unit(AU) with a large amount of pain information.By labeling the AU,the sample data set is obtained.Then the constraint set between BN conditional probabilities is obtained according to the experience of qualitative experts.The variable weight method is used to fuse the sample data set with the constraint extended parameter set to complete the parameter estimation of the BN model,and the BN reasoning method is used to realize the final recognition of pain expression.Experimental results show that compared with Probabilistic Latent Semantic Analysis(PLSA),Local Binary Volume Convolution Neural Network(LBVCNN) and other methods,this method can effectively improve the recognition performance of pain expressions and has higher recognition accuracy.

Key words: pain expression recognition, Constrained Local Neural Field (CLNF) model, Activity Unit (AU), Bayesian Network(BN) model, variable weight fusion, parameter estimation

中图分类号: 

• TP301.6