基于模糊聚类的“愤怒”表情细分方法研究

Research on“Angry” Expression Subdivision Method Based on Fuzzy Clustering

伴随情感计算和人机交互界面的快速发展,计算机的情感识别能力受到越来越多的关注。近年来针对面部表情识别存在很多方法,然而对于表情层次的细分研究却不多。目前网约车司机以及公交乘客的情绪失控情况无法被摄像头监控系统及时检测,该研究有助于此问题的解决。文中针对"愤怒"表情进行层次细分研究。首先通过RBF神经网络进行大类情绪识别,然后从已识别‘愤怒’情绪的多帧视频图像样本中选取出部分连续的图像样本。接着把选取出的连续样本进行融合聚类,确定初始聚类中心个数。最后通过自适应神经模糊推理系统(adaptive neural-based fuzzy inference system,ANFIS)对识别出的愤怒表情进行打分,分值越高愤怒程度越高。创新点在于情绪样本图片的选取上,基于人的情绪是有一定生成过程,选取同一个人"愤怒"时的连续多张图片作为样本。通过实验结果证明了该方法的有效性。

With the rapid development of emotional computing and human-computer interaction interface,more and more attention has been paid to the emotion recognition ability of computers.In recent years,there are many methods for facial expression recognition,but few subdivisions of expression level.At present,the emotional disorder of the network driver and the bus passenger cannot be detected by the camera monitoring system in time,so this research helps to solve this problem.The hierarchical analysis of“angry”expressions is carried out.First,a large class of emotion recognition is performed through the RBF neural network,and then a part of consecutive image samples are selected from the multi-frame video image samples that have been identified as“angry”emotions.Then,the selected continuous samples are subjected to fusion clustering to determine the number of initial cluster centers.Finally,the identified angry expressions are scored by the adaptive neural-based fuzzy inference system(ANFIS).The higher the score,the higher the degree of anger.The innovation lies in the selection of emotional sample pictures.Based on the human emotion,there is a certain generation process.The consecutive multiple pictures of the same person“angry”are selected as samples.According to the experiment,the method proposed is effective.