光流模值估計的微表情捕捉
發(fā)布時間:2019-05-15 11:27
【摘要】:采用力的加速度參量展開描述人臉表情的變化過程,直接反映變化速度,從而有效捕捉表情序列中由不完全肌肉運動所引起的微表情關鍵幀.利用Horn-Schunck(H-S)光流法對連續(xù)運動的人臉圖像序列提取運動目標的運動特征;通過光學應變張量算法,結合運動特征中的光流速度估計,推導出加速度參量;利用全局閾值算法對加速度模值和速度與張量模值作分類、比較,實現微表情圖像序列關鍵幀的提取.采用Oulu大學SMIC微表情數據庫中16個實驗對象的88個微表情片段作為實驗樣本,平均正確識別率可達80.7%,比僅利用光學張量算法的正確識別率高12.5%.實驗結果表明,所提出的加速度參量對微表情提取更具有效性.
[Abstract]:The acceleration parameter expansion of force describes the change process of facial expression and directly reflects the change speed, so as to effectively capture the microexpression key frame caused by incomplete muscle movement in the expression sequence. The Horn-Schunck (H 鈮,
本文編號:2477466
[Abstract]:The acceleration parameter expansion of force describes the change process of facial expression and directly reflects the change speed, so as to effectively capture the microexpression key frame caused by incomplete muscle movement in the expression sequence. The Horn-Schunck (H 鈮,
本文編號:2477466
本文鏈接:http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2477466.html