基于人脸多特征融合的精神分裂症自动识别  

Automatic recognition of schizophrenia based on fusion of facial multi-feature

作　　者：李雯[1] 郭湘 秦林雨 何凌[1] 郑秀娟 李元媛 LI Wen;GUO Xiang;QIN Lin-yu;HE Ling;ZHENG Xiu-juan;LI Yuan-yuan(College of Biomedical Engineering,Sichuan University,Chengdu 610065,China;College of Electrical Engineering,Sichuan University,Chengdu 610065,China;Huaxi Mental Health Center,Sichuan University,Chengdu 610041,China)

机构地区：[1]四川大学生物医学工程学院,四川成都610065 [2]四川大学电气工程学院,四川成都610065 [3]四川大学华西心理卫生中心,四川成都610041

出　　处：《计算机工程与设计》2023年第9期2761-2768,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(819013089);四川大学创新火花库基金项目(2018SCUH0002)。

摘　　要：针对精神分裂症诊断主观性强、缺乏客观辅助诊断指标的问题,提出基于人脸多特征融合的精神分裂症自动识别算法。提取人脸关键点,综合分析文本朗读过程中精神分裂症患者眼动、嘴部运动、头部姿态等非言语行为特征,构建融合特征并结合支持向量机进行自动分类识别。实验采集20位精神分裂症患者与20位正常对照组的文本朗读视频进行测试,其结果表明,提出方法在十折交叉验证中获得92.5%的分类正确率,算法实现了对精神分裂症的自动识别,能够为精神分裂症提供客观的辅助诊断指标。Aiming at the lack of objective diagnostic indicators in the clinical diagnosis of schizophrenia,an automatic schizophrenia recognition algorithm was proposed.Based on the extracted key points of the face,the nonverbal behaviors in the eye movement,mouth movement,and head posture of schizophrenic patients were analyzed.The above abnormal behavior characteristics were fused and combined with the support vector machine to realize the automatic detection of schizophrenia.Experimental data of this work were text reading videos recorded by 20 schizophrenic patients and 20 normal controls.Experimental results indicate that the classification accuracy of the proposed automatic schizophrenia recognition algorithm with 10-fold cross-validation is 92.5%.The proposed algorithm realizes the automatic detection of schizophrenia and it can be used as an objective auxiliary diagnostic method for schizophrenia.

关 键 词：精神分裂症 人脸关键点 非言语行为 融合特征 文本朗读 自动识别 辅助诊断 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]