雙影像視覺技術於人臉辨識之研究

Using Stereo Vision Technique for Face Recognition

蔡全益(chuan-yi tsai)

指導教授 ： 田方治

國立臺北科技大學/管理學院/工業工程與管理系所/碩士(2005年)

https://doi.org/10.6841/NTUT.2005.00115

摘要

在過去二十年裡，生物認證於安全維護方面被逐漸廣泛應用。尤其，以人臉為特徵而進行辨識，在生物認證領域裡，更是被積極研究著。本研究之主要目的為建構一套有效之人臉辨識系統，於特徵擷取方法上，採主成份分析(Principal Component Analysis, PCA)及區域自相關係數(Local Autocorrelation Coefficient, LAC)等兩種特徵擷取方法，並將此兩種特徵擷取方法進行整合，利用兩特徵擷取方法整合後，便會增加影像之特徵數，以提高人臉辨識結果。並利用二維影像之資訊加上三維影像之資訊，即加入影像之深度資訊，以突破僅採用二維影像之平面資訊，於辨識之瓶頸。最後於分類方法上，採用倒傳遞類神經網路(Back-Propagation Neural Network, BP)與歐氏距離(Euclidean Distance)等兩種分類方法，期能發展出一套有效之人臉辨識系統。本研究共對一百位受測者進行取像，每位受測者均取十三種不同之表情影像，實驗中，以表情一至表情七作為訓練影像，表情八至表情十三為測試影像，由實驗結果得知，將二維影像及三維影像之資訊進行整合，對辨識結果均能有效提昇；而PCA與區域自相關係數兩特徵擷取方法之結合，於歐氏距離上，對辨識結果並不能有效提昇，亦有可能造成反效果；但若改以倒傳遞類神經網路進行分類，對辨識率之提昇即有所助益；最後，將歐氏距離與倒傳遞類神經網路等兩分類方法相比較，得知倒傳遞類神經網路之分類效果優於歐氏距離之分類效果。

關鍵字

特徵擷取 ； 主成份分析 ； 區域自相關係數 ； 倒傳遞類神經網路 ； 歐氏距離

並列摘要

Biometric measurements received an increasing interest for security applications in the last two decades. In particularly, face recognition has been an active research in this area. The objective of this study is to develop an effective face recognition system that extracts both 2D and 3D face features to improve the recognition performance. The proposed method derives 3D face information using a designed stereo face system. Then, it retrieves 2D and 3D face features with Principle Component Analysis (PCA) and Local Autocorrelation Coefficient (LAC) respectively. Eventually, the information of features are fused and fed into a Euclidean-distance classifier and a Backpropagation neural network for recognition.
An experiment was conducted with 100 subjects. For each subject, thirteen stereo face images were taken with different expressions. Among them, the faces with expressions one to seven are used for training, and the rest of the expressions is used for testing. For the Euclidean-distance classifier, the proposed method does not improve the recognition result by combining the features derived from PCA with LAC; however, an improvement is observed when using the Back-Propagation Neural Network. In general, BP outperforms Euclidean distance in both 2D and 3D face recognition. Furthermore, the experimental results show that the proposed method effectively improves the recognition rate by combines the 2D with 3D face information.

並列關鍵字

Principal Component Analysis ； Local Autocorrelation ； Back Propagation Neural Network ； Euclidean Distance

參考文獻

[6] Shirai, Y. , Three-Dimensional Computer Vision, New York: Springer-Verlag,1987.

[8]Ackermann, F., “Digital image correlation-performance and potential application in photogrammetry,” Photogrammetric Record, Vol. 11, No.64, 1984, pp. 429-439.

[9] Akamatsus, S., Sasaki, T., fukamachi, H, and Suenaga, Y., “A robust face identification scheme-KL expansion of an invariant feature space,“ in SPIE Proc. : Intell. Robots and Computer Vision X: Algorithms and Techn, vol. 1607, 1991, pp.71-84.

[10] Bartlett, M.S., Movellan, J.R., and Sejnowski, T.J., “Face Recognition by Independent Component Analysis,” IEEE Trans. Neural Networks, vol. 13, no. 6, 2002, pp.1450-1464.

[12] Calitz, M.-F., and Ruether, H.,”Least absolute deviation (LAD) image matching,” ISPRS Journal of Photogrammetry and Remote Sensing ,Vol. 52, 1985, pp. 160-168.