自動臉部表情辨識在電腦視覺領域中仍然是具有挑戰性的問題。傳統的表情辨識通常使用靜態的單張影像,因此常受到不同照明、不同的視角及表情的細微差異所影響。為了克服這種不適定性問題,本研究旨在發展一套新方法,使用視訊中之動態臉部特徵來辨識五種特定的臉部表情(即無表情、生氣、高興、傷心及驚訝)。研究方法包括:(1) 臉部偵測;(2) 臉部特徵偵測;(3) 臉部特徵擷取;及(4) 分類模型。此外,整合有限狀態機針對臉部特徵之動態變化進行處理。初步研究結果顯示,本方法可以達到81.4%之整體辨識率。歸納而言,本研究提出一項自動臉部表情辨識的潛在解決方案。
Automatic facial expression recognition remains a challenging issue in the field of computer vision. Traditionally, recognition is typically based on the use of static single-image, thus is often subject to varying illumination, different view, and subtle difference of facial expressions. To overcome such an ill-posed problem, this study is aimed to develop a novel method for the recognition of five specific facial expressions (i.e., neutral, angry, happy, sad, and surprise) using dynamic facial features in video sequences. Our method includes: (1) face detection; (2) facial feature detection; (3) facial feature extraction; and (4) classification model. In addition, a finite-state-machine is integrated for processing dynamic changes of facial features. Our preliminary results indicated that our method could achieve an overall recognition of 81.4%. In summary, we have presented a potential solution to automatic facial expression recognition.