透過您的圖書館登入
IP:44.220.43.170
  • 學位論文

人臉表情辨識使用類神經網路與模糊推論應用於智慧型家電系統

Facial Expression Recognition by Using Artificial Neural Network and Fuzzy Inference Applied In Intelligent Household Appliance

指導教授 : 駱榮欽
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


臉部表情自動辨識在智慧型家電系統中是極為重要的一環。有許多的學者試圖利用自動化機械人辨識出人臉各種不同的情緒反應。雖然人類似乎在複雜的環境裡頭辨識出表情顯得十分容易,但是在機械視覺裡如何即時的執行人臉辨識卻是一大難題。而最困難的地方在於如何精確地分割並且擷取表情特徵。在這篇論文裡,我們發展了一套智慧型家電系統,可以偵測人臉和辨識表情,並且提出一套改良的色彩主動輪廓追蹤模型,使用臉部器官的色彩資訊來擷取人臉特徵。而我們所擷取的這些人臉特徵包括有眼睛、眉毛、嘴巴。除此之外,我們增加一個改進的外加能量到主動式輪廓追蹤模型裡,稱為Valley Energy,使得我們所得到的特徵輪廓能更加穩定。由於系統是設定在室內環境做人臉偵測,所以偵測人臉是直接在RGB色彩空間裡進行。而表情特徵擷取跟原始的主動輪廓追蹤模型比較起來,能夠更加精確找出器官的輪廓。我們所提出的演算法包含有五個步驟:第一,先從複雜背景中區分出可能的膚色區域;第二,找尋可能的人臉區域,並且使用最佳橢圓近似法來求得人臉的長軸、短軸、以及旋轉角度;第三,定位我們所要搜尋的臉部特徵;第四,使用改良式色彩主動輪廓追蹤模型擷取特徵;第五,使用類神經網路以及模糊決策來判別人的表情。

並列摘要


Automatic facial expression recognition applied in the intelligent household appliance system is a very important step. On this subject, most researchers attempt to recognize prototypic emotional expression. Although humans seem to recognize facial expression in cluttered scenes with relative ease, machine recognition in real-time is always a much more complex task. The major difficult issue is how to segment and extract human facial features precisely and then to recognize facial expression from these features. In the study, we also develop an intelligent household appliance system which can detect and recognize human facial expression. We propose an improved color active contour model (ICACM) that use color information in RGB color space from local facial organ to extract facial features. The features include the contours of eyes, eyebrows and mouth. To increase stability of the contour outline, we applied external energy named valley energy. From several experimental results, the method of facial feature extraction we proposed is more accurate than the original active contour model. The algorithm comprises five steps: First, discriminate skin color from background in RGB color space; second, search the candidates of the face region and fit in with best-fit ellipse; third, locate facial feature and set initial position of contour; fourth, use modified active contour model to extract facial features and feature points from the face candidate; fifth, recognize human facial expression by neural network and fuzzy decision system.

參考文獻


[1] P. Ekman, "Three Classes of Nonverbal Behavior", Aspects of Nonverbal Communication, Swets and Zeitlinger, 1980.
[2] Rein-Lien Hsu, Mohamed Abdel-Mottaleb and Anil K. Jain, "Face Detection in Color Images", IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 24, no. 5, 2002, pp. 696-706.
[4] M. Kass, A. Witkin, and D. Terzopoulos, "Snake: Active contour models", International Journal of Computer Vision, pp. 321-331.
[5] Kap-Ho Seo and Ju-Jang Lee, "Object Tracking Using Adaptive Color Snake Model", Proceedings of Advanced Intelligent Mechatronics. IEEE/ASME International Conference, vol. 2, 2003, pp. 1406-1410.
[6] S. R. Gunn and M. S. Nixon, "A Robust Snake Implementation: A dual Active Contour", IEEE Trans. Pattern Analysis and Machine Intelligence, 1997, vol. 19 no.1, pp. 63-68.

被引用紀錄


黃怡華(2016)。應用餘弦正規化與類神經網路於三維人臉辨識之研究〔碩士論文,逢甲大學〕。華藝線上圖書館。https://doi.org/10.6341%2ffcu.M0116393

延伸閱讀