Research On Recognition Method Of Fearful Emotion Based On The Facial Infrared Thermal Images

Emotion is a spontaneous,instantaneous physiological and psychological response,not controlled by subjective consciousness,but closely related to the biological basis.It is accompanied with human's cognitive and conscious process,and plays an important role in interpersonal communication.Emotion recognition is a complex research field which integrates many disciplines such as machine learning,cognitive science and psychological science.By extracting the relevant characteristics of human emotional expression and finding the statistical law,it is helpful to promote the harmonious human-computer interaction and has an important impact in the fields of psychological research,criminal polygraph,clinical medicine and so on.There is an inherent connection between human's emotion and physical signals.Although many researchers at home and abroad have achieved some results in the recognition or classification of emotion based on the physiological signals,the spontaneous reaction of the subject may affect the experimental results because the extraction process of physiological signals needs the sensors to contact with subjects directly.Fearful emotion is a response to the external stimuli of human.The generation of fearful emotion could lead to the change in the facial skin temperature.According to the principle of infrared thermal images reflecting the temperature distribution on the surface of objects,a non-contact measurement method based on infrared thermal images is proposed to recognize the specific fearful emotion.The main work accomplished in this paper includes:(1)Thirty-nine healthy subjects with no emotional disorder are selected to participate in the experiment.A video is used to induce the fearful emotion of the subjects.The entire experimental process is recorded by the RT513 infrared thermal camera.The above is the process of the acquisition of facial infrared thermal images.(2)A heat transfer model is simplified by curving fitting of exponential function,and the facial infrared thermal image is converted into the blood perfusion pseudo-color image to find the regions of interest.Then,the experimental data can be processed and analyzed in the following work.(3)The temperature values of the region of the interest(forehead region)of subjects are converted into the blood perfusion values by the heat transfer model.And the blood perfusion change curves are obtained.Then,features of the blood perfusion change curves are extracted(slope,confidence coefficient,mean value,and standard deviation),and the correlation between the features and the self-assessment score of the fearful emotion is analyzed using Spearman correlation coefficient.Finally,the standard deviation(with a threshold of 0.14g·m-2·s-1)which is highly related to the self-assessment score is applied to recognize the fearful emotion of the subject.And,the threshold of the standard deviation is obtained by K-means clustering algorithm.Eliminating the wrong experimental data,the proposed method is demonstrated to be satisfactory and reliable with an accuracy of 85.7%for all the 28 tested subjects.(4)In order to verify the validity of the proposed method of recognition of the fearful emotion,a set of test experiments is added and the facial infrared thermal images of 17 subjects are collected.The facial infrared thermal images of the test set are processed and analyzed in the same way,and the region of interest and the standard deviation threshold are verified obtained by the sample set.Excluding erroneous data of two subjects,15 samples of the test set are tested with an accuracy of 86.7%.Meanwhile,the blood perfusion curves of the subjects in the sample set and the test set are contrasted and analyzed with a result of the characteristics being consistent.There is an obvious decrease in the blood perfusion of forehead region in the presence of fearful emotion,which is consistent with observations of previous studies.