模式识别国家重点室-VideoLecturesNET.ppt

Facial Expression Analysis (CVPR’0708) Static Image based Methods Video based Methods A nature facial expression evolves over time from the onset, the apex, to the offset. The key issue is how to describe the dynamics of the expression? 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Facial Expression 1. 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Video based Methods The time-resolution issue Facial Expression Analysis (CVPR’ 08) Facial Expression 2. 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Our Solution Harr-like feature representation Gaussian based temporal pattern models Facial Expression Analysis (CVPR’ 08) Facial Expression 3. 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Our Solution Dynamic binary patterns Adaboost based classifier Facial Expression Analysis (CVPR’ 08) Facial Expression 4. 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Facial Expression Analysis (CVPR’ 08) Facial Expression 5. 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Facial Expression Analysis (CVPR’ 08) Facial Expression 6. Face Alignment (ICCV’07) Active Shape Model (ASM) 模式识别国家重点实验室 中国科学院自动化研究所 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences Face Alignment 1. The original ASM PCA based global shape model Local texture models for every shape point The issues Points are independent in search Global shape model is linear Component based Framework Local model: PCA + Nearest neighborhood constrain between po