以主成份分析為基礎之嬰兒表情辨識系統

An Infant Facial Expression Recognition System Based on Principal Component Analysis

黃律嘉(Liu-Jia Huang)

指導教授 ： 方瓊瑤

國立臺灣師範大學/理學院/資訊工程研究所/碩士(2011年)

摘要

本論文提出一套具有辨識嬰兒表情功能的視覺式監控系統。因為年幼的嬰兒還不會用言語來表達自身的需求，我們只能透過嬰兒臉上的表情來了解嬰兒是否需要幫助。此外全天候照顧嬰兒需要花費許多精力，本研究擬建立智慧型表情辨識輔助系統，藉由判別嬰兒的表情，隨時告知照顧者嬰兒的現況，系統如果偵測到嬰兒哭泣或吐奶時，能及時發出警告提醒照顧者檢視嬰兒的情況，以求降低嬰兒意外的發生。
本研究將攝影機架設在嬰兒床前方，拍攝嬰兒的臉部影像，輸入至系統作分析。其中系統可分兩個子系統：嬰兒臉部偵測以及嬰兒表情辨識。嬰兒臉部偵測是利用Locus model定義的膚色範圍找出影像中的膚色像素，並利用connected component的技術找出最大的膚色區塊，經由區塊分析可找到嬰兒臉部區域，完成嬰兒臉部定位後。即可擷取出嬰兒臉部區域影像。
在臉部特徵擷取部分，本研究採用主成份分析(principal component analysis)的方法，取出具有代表性的臉部特徵，並將訓練影像的特徵向量與投影矩陣存入資料庫中。之後只要將臉部表情的連續影像輸入至系統，並和資料庫中的投影矩陣相乘就能得到單張影像的特徵向量，與資料庫中的各表情影像做比對後，可得出該影像的表情類別。
因為單張表情辨識的正確率無法提高，我們考慮連續影像的表情發生機率，直到有某種表情類別的發生機率超過系統預設的臨界值，系統才將辨識結果輸出，由實驗結果證明本系統是穩定且有效率的。

並列摘要

This thesis presents an automatic infant facial expression recognition system. Since infants are too young to say what they need, we can only guess the infants’ needs by their facial expressions. Moreover, to take care of the infants from day and night is a hard work. An infant facial expression recognition system can help baby sitters to monitor the infants to decrease their load and to avoid the infant accidents.
We set up the camera on the infant’s crib and obtain the infant’s face images to input into the system. The system can be divided into two main stages: infant face localization and facial expression recognition. At the infant face localization stage, the system applies the locus model to detect the skin color pixels from the input images. After that, the system uses the connect component technology to find the biggest skin color region which is regarded as the infant’s face.
At the facial expression recognition stage, the system analyzes and classifies infant’s face images according to their facial expressions. The system classifies the facial expressions based on the feature vectors extracted by the principal component analysis method (PCA). Once a testing image input into the system, the feature vector is extracted by PCA and is classified by the nearest Euclidean distance method.
Since the error rate of individual image cannot be reduced, the system considers the evidence supported by successive images and outputs the final facial expression if the evidence is enough. The experimental results show that the proposed system is robust and efficient.

參考文獻

[黃04] 黃泰祥，「具備人臉追蹤與辨識功能的一個智慧型數位監視系統」，私立中原大學碩士論文，2004。

[Kak07] P. Kakumanu, S. Makrogiannis, and N.Bourbakis, “A survey of skin-color modeling and detection methods,” The Journal of the Pattern Recognition Society, pp.1106-1122, 2007.

[Zha09] P. Zhang, “A Video-based face detection and recognition system using cascade face verification modules,” Proceedings of the Applied Imagery Pattern Recognition Workshop (AIPR 2008), Washington, DC., 2008.

[Phu05] S. L. Phung, A. Bouzerdoum, and D. Chai, “Skin segmentation using color pixel classification: analysis and comparison,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 27, no. 1, pp.148-154, 2005.

[Tom94] C. Tomasi and J. Shi, “Good features to track,” Proceedings of the 9th IEEE Computer Society Conference on Computer Vision and Pattern Recognition, California, pp. 593-600, 1994.