運用人臉特徵及類神經網路於人臉表情辨識系統

A Facial Expression Recognition System Using Facial Feature Extraction and Artificial Neural Networks

江妍瑤(Yan-Yau Jiang) ； 張元翔(Yuan-Hsiang Chang) ； 李忠驊(Chung-Hua Li,) ； 楊靜杰(Jin-Jie Yang)

《先進工程學刊》 3卷2期 (2008/04) Pp. 153-159

https://doi.org/10.29948/JAE.200804.0008

摘要

臉部表情是人類表示情緒狀態的重要特徵。本研究整合臉部特徵（例如：眉毛、眼睛等）位置變化的情況來定義表情，發展出一套「臉部表情辨識」系統，從臉部影像辨別各種表情（例如：無表情、微笑、生氣、悲傷和快樂）。首先利用影像處理技術自動擷取臉部特徵，將其對一參考軸正規化，且將這些經正規化特徵作為類神經網路之輸入，而類神經網路之輸出即為表情辨識結果。本系統使用本研究室建立之560個臉部影像的表情資料庫，其中280個為訓練樣本，280個為測試樣本。另外並採用JAFFE表情資料庫作為測試樣本。研究結果顯示，本系統可達到約88%的辨識率，JAFFE表情資料庫辨識率約87%。實驗結果顯示本系統對於臉部表情可達到初步成功之辨識結果。

關鍵字

類神經網路 ； 表情辨識 ； 特徵擷取 ； 影像處理

並列摘要

Facial expression is an important human characteristic that indicates a person’s motional state. In this paper, facial expression is defined as the combined results of position changes in facial features (e.g., eyebrows, eyes, etc.). We propose a facial expression recognition system to distinguish various facial expressions (i.e., neutral, smile, angry, sad, or happy) using face images. Facial features are first automatically extracted using image processing techniques and normalized with respect to a central axis. These normalized features are then used as inputs to an artificial neural network (ANN) and the ANN outputs are used to indicate the potential facial expression. During the experiment, 560 face images were collected in our laboratory, among which, 280 images were formed as the training set, and the remaining 280 images were used for testing. In addition, the JAFFE database was used as testing samples. Our results demonstrated that the system has achieved a reasonable detection rate of approximately 88% in our database and 87% in the JAFFE database. In summary, the system was shown to recognize given facial expressions with preliminary success.

並列關鍵字

artificial neural network ； facial expression ； feature extraction ； image processing