A Face Tracking and Recognition Using AdaBoost and PCA Algorithm

10.29768/JNTUT.201106.0002

駱榮欽(Rong-Chin Lo)；許秀貞(Hsiu-Chen Hsu)

人臉辨識 ； 主成份分析 ； AdaBoost ； 影像處理 ； Face Recognition ； PCA ； AdaBoost ； Image Procession

臺北科技大學學報

44卷1期（2011 / 06 / 01）

21 - 33

英文

本研究的目地是發展即時人臉辨識於智慧家庭使用之無線鏡面顯示系統中之嵌入式系統，藉由家庭成員的正面影像做為訓練樣本，首先利用積分影像結合Haar函數及AdaBoost演算法，由灰階化影像中將臉及臉部區域部份（眼、鼻、嘴）偵測出，再以矩型框出後，加以正規化取得的臉與臉部區域部份影像。影像定位後，以主成份分析(PCA)法取出臉及區域臉部的影像特徵向量，在特徵空間(Eigenspace)中建立各特徵資料庫。當系統執行即時辨識時，將測試影像之臉及區域臉部的特徵，投影在特徵空間中與各特徵資料庫比對。為避免光照變化的影響，利用Retinex法強化影像特徵；為避免歪斜的影響，利用眼線水平校正；為減少表情的影響，依其重要性調整估算權重，以提高辨識率。由實驗結果，所提之方法，確實有達到即時之辨識效果，辨識率可達96%以上。

In this paper, we present a method of real-time face recognition for embedded system of wireless mirror display system at smart home. To this end, the proposed approach takes advantages of the improved integral image concept and a human facial detecting training method of Adaptive Boosting (Adaboost). The facial recognition approach combines the features of whole-face and facial areas (eyes, nose and mouth) by using Principal Component Analysis (PCA) method. All the compared features are weighted to determine who the test image is. The paper also presents an analysis and evaluation regarding the influence of the variation of facial illumination, pose and expressions of the input images. To decrease the influence of illumination variation, Retinex algorithm is adopted. To adjust the incline pose, the eye-line is detected to adjust face to horizontal level. To reduce the influence of expression, the weights of those features are decreased according to the affect by facial expression. The experimental results, showing the availability and the usability of the proposed approach with recognition rate up to 96%, are included.