臉部表情在人類的認知行為中是不可或缺的,也可以藉由表情顯現出人類當時的情緒。有鑑於此,本研究是以臉部表情為導向,發展出一套「臉部表情辨識」系統,讓電腦對使用者的情緒做出相對的回應,以達到人機溝通的目的。 本研究是利用數位相機對人臉進行拍攝,並將影像傳輸至電腦中進行處理。接著利用臉部特徵(Facial Feature)進行特徵抽取(Feature Extraction),再根據這些特徵的位置計算出分析表情時所需要的特徵點座標及特徵值,之後開始收集資料,分析各種表情下特徵點和特徵值的變化情形,最後將分析的結果加入倒傳遞類神經網路系統中,該系統根據此結果對即時的人臉影像進行表情判別。 在研究結果部分,我們以28位受測者(14位為訓練樣本,14位為測試樣本)做為資料庫,每一人各有5種表情(Normal、Smile、Angry、Sad、Happy),每個表情各做4次,而結果顯示整體的平均辨識率都達85%左右,這表示本系統對於不同的受測者能夠做出良好的判斷也證明該網路趨於穩定。
Facial expression is an important human cognitive behavior that generally reflect the person’s mood at the time. The objective of this research is to develop a “facial expression recognition” system using the facial expression as key features to produce relevant response to the person’s mood, with an attempt to achieve an effective human-computer interaction. This research utilizes a digital camera to capture face images for computer processes. Facial features are identified using the feature extraction method. To characterize facial expression, facial feature locations(or coordinates) are identified and collected for further processes. Then, these location and their corresponding relation are analyzed and used as inputs to an artificial neural network(ANN). Finally, outputs of the ANN are used to determine the person’s facial expression. A total of 28 persons was invited to participate in this research. Among which, 14 were used as the training set, the remaining 14 were used as the testing set. Each person was asked to repeat 4 times for 5 different facial expression (i.e., normal, smile, angry, sad, and happy). The results demonstrated that our system has achieved the detection rate of 85%. In summary, our system is shown to successfully recognize the given facial expressions.