| 摘要: | 由國際生物辨識組織的研究報告得知，利用人類生理特徵作為識別個人身分的生物辨識技術，將成為未來安全機制的主流。而在各種生物辨識方法中，人臉辨識在使用上最為簡便與人性化，若將其結合視訊監控系統，更可主動對範圍內的行人進行身分的比對與確認，也因此人臉辨識成為目前最熱門的研究主題之一。 在本文中，研究的重心是集中於以人臉外表特徵(appearance feature)為基礎之特徵擷取法上，其中最具代表性的方法是主成份分析法與獨立成份分析法。由於傳統的計算架構通常將整張人臉影像視為單一樣式(pattern)，而人臉局部特徵的統計資訊則有被忽略的可能，故後來有學者提出以次樣式(sub-pattern)為基礎的計算架構，試圖改善傳統方法之缺點。然而，我們進行一系列的實驗後發現，在某些情況下，次樣式特徵擷取架構的辨識率反而低於傳統方式。針對此一現象，本文進行深入的分析與探討，並提出合理的物理解釋。最後，為了同時考慮人臉整體與局部特徵的統計資訊，本文發展出一混合型特徵擷取架構，經由三個人臉資料庫測試的結果顯示，此架構有更佳的辨識性能。 此外，許多研究文獻指出：同一人的影像因受到光線變化所造成的改變，通常大於不同人影像間的差異，因此，光線變異對辨識性能的影響可見一斑。為了改善這個問題，本文提出一個光線補償方法，其作法是：先對影像中的局部區域進行直方圖等化，以強化局部特徵，接著透過鏡射與改良式影像平均的技術，融合左右兩邊的人臉資訊，以達到光線補償的效果。此方法具有以下特點：(a)架構簡單，容易實現；(b)能強化人臉主要特徵，並將非人臉主要特徵標準化；(c)能直接應用於單張影像，而無須預知光源方向，或使用多張影像進行訓練。 In a variety of biometric techniques, face recognition is a natural, non-intrusive, and user-friendly scheme. Especially, face recognition has the great advantage of being able to in places with large concourse of unaware visitors. After the Sept. 11, 2001 terrorist attacks, this active property of recognizing identity is not only garnering much more attention in both academia and industry, but also pushes face recognition turn into one of the most popular research topics. In this dissertation, we mainly focus our attention on the investigation of appearance- based face recognition methods. Since the conventional appearance-based methods (e.g. Eigenfaces and Fisherfaces techniques) usually consider whole face image as a single pattern, some local statistical information of face image may be ignored. In order to emphasize the local facial features, some researches proposed their approaches which try to find the local appearance features by sub-pattern technique. Nevertheless, after a serious of experiments, we found that in some case the recognition accuracy of the sub-pattern based method is lower than that of the corresponding conventional method. In the former half of this dissertation, we presented our opinions to reasonably explain what causes this result. Then, by simultaneously considering global and local information of face images, we developed a novel hybrid approach for face recognition. The experimental results show that the proposed hybrid approach has better recognition performance than that obtained using other traditional methods. In addition, some literatures pointed out that the intra-person variations caused by illumination change are often larger than the inter-person differences. Consequently, illumination variation is key factor that can significantly affect system face recognition performances. In the latter half of this dissertation, we proposed a novel shadow compensation method for dealing with this problem. Among others, this method has several advantages: (1) it is very simple so it is easily implemented in a real-time face recognition system; (2) it is able to reinforce key facial features and to standardize other parts of the face; (3) it can apply directly to single face image without any prior information of light source direction. |
