相較於其他生物辨識方法，人臉辨識在使用上最為簡便與人性化，若將其與視訊監控系統結合，更可主動地對範圍內的行人進行身份的比對與確認，此一特性使得人臉辨識的相關研究，成為目前學術界與產業界最熱門的研究主題之一。然而，影響人臉辨識性能最大的因素，在於訓練影像與測試影像之間的不匹配，其原因大致可歸納成兩類：環境導致的不匹配與人臉本身的不匹配。傳統以統計方法為基礎的辨識方式，通常將整張人臉影像視為單一樣式(pattern)來進行特徵擷取與辨識，而人臉局部特徴的統計資訊則有被忽略的可能性。故本研究計畫的主要目的是發展以次樣式為基礎之人臉辨識架構，以期能改善傳統方法可能忽略局部特徵統計資訊的缺點，進而提高人臉辨識系統的強健性與辨識性能。 ; The use of face recognition is very simple and convenient for user in comparison with other biometrics technology. The face recognition technology can be combined with video surveillance system, and then we can actively recognize identification of every person in the region of surveillance. Hence, the face recognition technology becomes one of the most popular research subjects. Though many face recognition systems have successfully been applied in well-controlled environments, the task of robust face recognition for uncontrolled environments is still difficult. The most important factor that seriously influences the performance of the recognition system is large variation between training images and testing images. The reasons of causing large variation can roughly be summed up into two kinds: the illumination variation problem and the pose variation problem. Since the traditional statistical methods for face recognition only considered the global statistics information among the all whole patterns. Some local statistics information may not be emphasized. Therefore, the sub-pattern based face recognition architecture was proposed in this project for improving the performance of the face recognition system. ; 研究期間 9708 ~ 9807
