提供一個可靠且有效率的人臉辨識系統需克服許多問題，其中包含了人臉影像受到光線、表情或頭部姿勢變化等，這些因素將會造成辨識率的下降，因此所採用之系統在面對影像的變化時，應具備較高之抵抗能力。此外，若要將系統架構能應用在即時辨識上，即要考慮到系統的計算時間必須是快速且有效率的。 基於上述目的之下，本篇論文提出了一個架構，可同時克服影像中具有光線變化與臉部表情變化的問題，並以最少的訓練影像張數、較少的計算時間，達到最佳的辨識率。本系統先使用Retinex演算法降低光線的影響，再使用主動式外觀模型（Active Appearance Model, AAM），擷取出五官特徵，建立component-based的辨識系統，再利用Support Vector Machine (SVM)以嘴巴特徵來建立表情分類的model，並依據SVM的判斷結果，削弱受到表情影響特徵的權重值，藉以降低表情變化對辨識率帶來的影響。 而實驗結果顯示，本篇論文所提出的架構，在每個人只訓練單張影像下，能克服光線變化與表情變化的問題，提升了系統的辨識率。 Most face recognition methods assume either constant lighting condition or natural facial expressions and hence can not deal with both kinds of variations simultaneously. The constraint has to be alleviated in a reliable and practical face recognition system. In order to resolve the aforementioned problem, we present a component-based face recognition system which can deal with both illumination and facial expression variations with the using of only one training sample image per class. In our work, retinex algorithm is firstly adopted to decrease the influence of illumination variation. Then, active appearance model (AAM) is employed to extract facial features to establish the proposed face recognition system. Next, support vector machine (SVM) is utilized to distinguish the variations of facial expressions by using the mouth features. To equip with the capability of insensitivity to expressions, the proposed system decreases the weights of these features which are affected by facial expressions. Finally, the recognition part combines the global feature and local features to generate the recognition result. Experimental results demonstrate that the proposed component-based face recognition system can indeed improve the performance when the images are under different illumination and facial expression variations.
