近年來Human computer interaction (HCI),也就是所謂的人機介面,在很多應用領域上面愈來愈重要。一些傳統的I/O介面,例如鍵盤、滑鼠、麥克風,已經不能符合新時代的需求。新穎的人機介面與傳統I/O的有著不盡相同的著眼點。最重要的不同在於,新型的人機介面將更考慮使用者的「人性」。「人臉」的資訊在智慧型影像人機介面的重要性是可以預知的。人臉辨識在最近幾年變得非常熱門。這個領域非常吸引學者去研究。其中涵括的領域不乏有電腦視覺(computer vision)、圖形辨識(pattern recognition)、機器學習(machine learning)、生物統計學(biometrics)。人臉辨識在生物檢測上的運用是一個非常令人期待的技術。但是在一個實際應用的系統中,最重要的前提是必須知道人臉的位置。 本篇論文主要在架構一連串與人臉有關的系統,目的在達成動態人臉辨識的能力。我們利用許多前者所提出的方法,架構出一個新架構的動態人臉辨識流程。由於有些方法在應用上的方式不同,在架構系統時會遇到需多問題。本篇論文除了提出解決之道外,也提出適用本系統的新方法。這個新架構不單只停留在理論階段。我們利用硬體和軟體實現整個系統,並使系統有能力在新環境下可以迅速重新初始化,藉此希望可以達成真正實際運用的目標。
Human computer interaction (HCI) has placed more and more importance on many applications. The traditional I/O (input/output), such as keyboard, mouse, and microphone, has no longer kept up with the new times. The novel HCI does not rely on the traditional I/O. It attends to be more normal human feeling. The human faces are very essential to the intelligent vision-based human computer interaction. Face recognition becomes very popular recently. It attracts the researches very much. It includes the fields in computer vision, pattern recognition, machine learning and biometrics for years. Face recognition is an important biometric identification technology. However, it bases on the conditions that we have already known at which the position of the face is. In other words, to implement a successful facial analysis system, the most important step is to detect face automatically, accurately, and rapidly. There are many topics of face jobs such as pose estimation, facial feature extraction, facial expression, sex distinction, face detection, face recognition, and face tracking. Each topic has its purposes and applications. In the dissertation, the system has a series of face jobs combined. They are face tracking, face detection, and face recognition. There are many difficulties in the sequent jobs. The new architecture applies face detection and face recognition in the still images to the dynamic tracking. Although these topics have been studied for years, there are still many problems in a real application. There are many applicable methods which are modified and combined with each other.