影像處理技術與類神經網路於表情辨識之應用

Facial Expression Recognition Using Image Processing Techniques and Neural Networks

吳佳穎(Chia-Ying Wu)

指導教授 ： 李錫捷

元智大學/資訊學院/資訊管理學系/碩士(2012年)

摘要

人臉表情在人與人互動中隱含著重要的資訊，因此，於日常生活中扮演著不可或缺的角色。近幾年人臉表情辨識的研究領域已被廣泛討論，在傳統的表情辨識系統中，大部分重於人臉偵測與特徵點定位，但人臉表情除了這些特徵外，亦會伴隨著某些肌肉線條。因此，本論文除了以傳統表情辨識系統的方法外，亦增加偵測臉部的肌肉線條紋理，如眉頭紋、額頭紋與下巴紋理等特徵進行表情辨識，故研究內容包含人臉偵測、特徵擷取、紋理偵測與表情辨識四個部分。
首先藉由人臉偵測技術於影像中擷取人臉區域，並於人臉區域擷取眼睛、鼻孔、嘴唇與眉毛的特徵點，再根據人臉五官相依關係擷取特徵紋理的位置，最後利用這些特徵點與紋理建立出一組特徵向量，以類神經網路SimNet來判斷使用者的臉部表情。
在實驗與討論中，本論文以TFEID表情資料庫與FEI Face人臉資料庫分別作表情辨識與人臉辨識，由實驗顯示，以個人化的表情辨識實驗TFEID資料庫可達96.2%，全體成員的表情辨識為92.8%。以TFEID為樣本的人臉辨識為97.4%，
以FEI Face為樣本的人臉辨識可達87.0%。

關鍵字

表情辨識 ； 紋理偵測 ； 特徵區域 ； 特徵擷取 ； 人臉辨識

並列摘要

In our daily life, facial expression, which implicitly contains important information, involve a certain role when people interacting with each other. These years, Human Facial Expression Recognition has been researched for a period. In traditional facial recognition method, most of them focused on face detection and feature point. They recognize facial expression with obvious feature area. However, except these features, human face also accompanies with muscle streaks. As a result, in addition to use traditional method as a face feature extraction mechanism, the paper also adds facial muscle streak, for example nasolabial folds and frown lines, as a recognition condition. Consequently, the research content contains Face Detection, Feature Extraction, Contour Detection and Facial Expression Recognition four parts.
First, use traditional face detection to extract face area from original image. Then extracting eyes, mouth and eyebrow outlines’ position form face area. Afterward, extracting important contours from different feature areas, and use them to create a set of feature vector. Then, these vectors can be processed with neural network and determine user’s facial expression at last.
In experiment and discussion, the paper using TFEID and FEI database on expression recognition and face recognition. The experiment result shows, 96.2% of TFEID database can be recognized in personalizing expression recognition experiment; full member recognition rate is 92.8%. In face recognition, 97.4% of TFEID sample can be recognized; 87.0% of FEI Face sample can be recognized in average.

並列關鍵字

Facial Expression Detection ； Contour Detection ； Feature Area ； Feature Extraction ； Face Recognition

參考文獻

[27] 黃欣慧, "影像處理技術應用於人臉膚質特徵之研究," 碩士, 資訊管理學系, 元智大學, 桃園縣, 2006.

[8] S. K. Singh, D. Chauhan, M. Vatsa, and R. Singh, "A robust skin color based face detection algorithm," Tamkang Journal of Science and Engineering, vol. 6, pp. 227-234, 2003.

[25] R. C. Gonzalez and R. E. Woods, "Digital image processing second edition," Beijing: Publishing House of Electronics Industry, 2002.

[1] A. Mehrabian and M. Wiener, " Decoding of Inconsistent Communications," Journal of Personality and Social Psychology, vol. 6, pp. 109-114, 1967.

[2] C. Darwin, The expression of the emotions in man and animals: D. Appleton and company, 1874.