基于深度视频分析的面瘫分级方法  

Classification of Facial Paralysis Based on Deep Video Analysis

作　　者：段群 郭新明 黄素萍 谢飞 DUAN Qun;GUO Xinming;HUANG Suping;XIE Fei(School of Computer Science, Xianyang Normal University, Xianyang 712000, China;School of Computer Science, Northwest Polytechnical University, Xi’an 710129, China;Academy of Advanced Interdisciplinary, Xidian University, Xi’an 710068, China)

机构地区：[1]咸阳师范学院计算机学院,陕西咸阳712000 [2]西北工业大学计算机学院,陕西西安710129 [3]西安电子科技大学前沿交叉研究院,陕西西安710068

出　　处：《微型电脑应用》2021年第12期23-25,32,共4页Microcomputer Applications

基　　金：陕西省教育厅服务地方科学研究计划项目(19JC041)。

摘　　要：针对面瘫患者诊断视频中面部运动特征的复杂性,设计能够学习更本质面部特征的深层非线性网络结构,结合可见光和深度视频数据更全面地分析面部运动细节特征。该方法主要使用具有多重卷积结构的网络模型和LSTM网络,分析患者与正常人面部运动之间的差异,完成真假面瘫识别;提取面部运动过程中的时序形变特征和纹理变化特征,并进行特征融合,实现面瘫分级评估问题;深入分析患者面部运动在可见光视频和深度视频中所体现出的细微变化特征,使用融合多源视频信息的卷积神经网络模型对面瘫等级进行细粒度评估。Aiming at the complexity of facial motion features in the diagnosis video of patients with facial paralysis,a deep nonlinear network structure which can learn more essential facial features is designed.Combined with visible light and depth video data,facial motion details are analyzed more comprehensively.This method mainly uses the network model with multiple convolution structure and LSTM network to analyze the difference between the facial movement of patients and normal people to complete the recognition of true and false facial paralysis;extracts the temporal deformation features and texture change features in the process of facial movement,and carries out feature fusion to realize the classification evaluation of facial paralysis;deeply analyzes the visual and depth video features of patients’facial movement.The convolution neural network model based on multi-source video information is used to evaluate the facial paralysis level.

关 键 词：深度视频分析 面瘫分级 主动外观模型 多重卷积神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]