Research Into Assessment Of Facial Paralysis Based On Active Appearance Model

The majority research into human faces in computer vision and patternrecognition is concerned with normal people, i.e. their faces are not affected by anydisease and appear symmetric or approximately symmetric. However, a great numberof people in the world are suffering from facial paralysis. It is caused when the facialnerve, which sends nerve impulses to the muscles of the face, loses its function. Facialparalysis results in significant psychological and functional disability from theimpairment of facial expression, communication and eye protection. For example, itusually affects one side of the face, causing a drooping mouth, drooling, andexcessive tearing from one eye. There are so many people who suffer from facialparalysis. Thus, to develop an objective and fully automatic facial nerve evaluationsystem has important theoretical significance and practical value. Our aim is toprovide such a system.In this paper, we provide a new approach which is based on the ActiveAppearance Model for accurately and objectively estimating the degree of facialparalysis. First, we use the AAM algorithm to calibrate the facial feature points, locatethe key points of facial features, and then estimate the degree of asymmetry inpatients with left and right side of my face. For each patient, we collect five actionimages: close eyes, raise eyebrows, wrinkle nose, plump cheeks and open mouth. Wedid lots of experiments and worked out a weight value for each action to calculate thefinal score. Finally, we fit the experiments’ results with the experts’ results. The higherthe score is, the severer the prevalence of patients is; and the smaller the score is, theslighter the prevalence of patients is. Experimental results show that the resultsobtained by the above method consistent with the diagnosis of clinical experts and thesystem certain practical value.