Research On Evaluating Method Of Paralysed Faces Based On Facial Pattern Features

Facial paralysis is multiple,repetitive and difficult to be cured.In addition to suffering from facial distortions,imbalances and other physical injuries,patients also suffer a heavy psychological blow and lack confidence in communicating with the outside world.Long-term rehabilitation training can effectively improve the patient’s paralyzed state.However,in the current facial rehabilitation training methods,training in front of a mirror and EMG feedback training lack quantitative training feedback,poor feedback effect and occupy a large amount of medical resources,which cannot be promoted and applied.In collaboration with Shanghai Ninth People’s Hospital,this paper proposes the static and dynamic quantitative evaluation methods of facial muscles to guide the rehabilitation training process and establish a standardized,convenient and efficient diagnosis and treatment system.In this paper,the HOG-Adaboost classifiers are adopted to detect paralyzed faces and the cascade regression trees are adopted to achieve face alignment of 68 feature points.The shape asymmetry indexes and the position asymmetry indexes of eyebrows,eyes,the nose,and the mouth are respectively determined based on precise coordinates of 68 facial feature points.A static evaluation method of paralyzed faces is established by quantifying the asymmetry of various parts of the face,which can recognize facial feature points automatically.Secondly,an annotated dataset of facial nasolabial regions and nasolabial folds are established.The accurate location of the nasolabial folds is achieved by adoptingnasolabial regions detection based on Faster-RCNN and adopting shape extraction for nasolabial folds based on GCN.The asymmetry of nasolabial folds are quantified by comparing the length,area,angle and coordinates of nasolabial folds.A static evaluation method of paralyzed faces based on the nasolabial fold is established for the first time.After selecting the axis of symmetry during the movement of paralyzed faces,facial voluntary movement and facial synkinesis are evaluated for analysis.The dynamic evaluation method of paralyzed faces is established.The experiments confirmed that evaluation system based on facial pattern features proposed in this paper achieved the same results as the Sunnybrook facial scoring system,which proved the scientificity and reliability of this system.