基于人脸分析技术的中医阴阳五行人分类研究

Research on yin and yang and five-element characteristics classification in traditional Chinese medicine by face analysis technology

目的:应用人脸检测技术,探索中医五行人的机器辨识功能。方法:基于主动外观模型(AAM)人脸特征点识别技术,完成人脸关键点自动识别;依据中医望诊经验以及相关经典论述,分解人脸图像信息,形成93项人脸形态单元与30项色彩单元,完成对中医医师望诊信息的数字化精确表达。通过对前期采集的2000例标准化的人脸的五行属性的人工分类,并进入机器分析,得到五行样本间差异性的数据指标,通过总结差异性的模式数据,设计完成基于人脸图像的机器预判方案。在进一步的准确性评价试验中,按照规范化的方法采集方法,完成了68名受试者样本人脸信息的规范化采集,并接受中医医师的五行辨识。结果:比较中医医师与机器分析得到的五行分类结果,发现二者判别的结论的一致性达到66.18%。但在五行中每一行的辨识精确度方面,土行人脸样本的准确性仍有待提升。结论:五行人智能识别系统可以实现对五行人的分类预判。

Objective:By applying face detection technology,this paper explores the ability of AI technic in fiveelement characteristics classification based on traditional Chinese medicine(TCM)principle.Methods:Based on the active appearance model(AAM)fitting method,key landmark in facial image would be automatically located.According to the TCM theory and experience of TCM physicians in face inspection,we digitalized the details of facial image,and concluded 93 facial shape and 30 facial colors indexes to describe facial details and describe the facial observation conclusions of TCM physician in digital expression.By manually classify former collected 2000 standard facial image,then digitalized the grouped sample by the designed method,those indexes of significant difference among the five elements samples group were obtained.Based on the dataset of those indexes,a method to automatically facial identify system was built.In further accuracy evaluation test,68 subjects were enrolled and completed the standardized facial image taken procedure,and accepted the dialectic analysis by TCM physicians as well.Results:by comparing the five elements classification results obtained by TCM doctors and machine analysis,it was found that the consistency of the two conclusions was reached to 66.18%.However,in terms of the accuracy identification of each one of the five elements,the accuracy in predict Tu element face samples still needs to be improved.Conclusion:the automatic fiveelement facial recognition system of can realize the classification prediction in certain accuracy level.