人臉偵測之研究

Abstract

近年來，數位安全監控技術的普遍發展，使得完全自動化的人臉系統 (fully automated face system) 越來越受到重視。人臉偵測是自動化人臉系統的基石，任何人臉應用系統都需要有一個強固的人臉偵測方法，才能達到預期的效果，因此人臉偵測的研究與分析便顯得重要而且迫切。雖然人臉偵測方法的研究已經超過 20 年，也有許多成果已經發表。可是對於多變化環境下的人臉偵測，特別是針對不同人臉角度 、不同光源方向與有顏色光源的人臉位置偵測方法，依然是一項十分具挑戰性的研究領域。 一般人臉偵測方法的主要架構大致是由下列幾項要素組成: 人臉候選位置的判定、臉部特徵的擷取與人臉位置的偵測。可是由於多變化的環境，導致人臉偵測的困難。例如在研判人臉可能出現的位置時，就可能受取像時的環境條件所影響，有顏色的光源與不同角度的側面光源，如不加以考量就會影響人臉候選位置的判定。另外由於攝影機與人臉之間的取像角度，使得人臉在影像中所呈現的姿勢可能為正面、接近正面、半側面、側面等，一些重要的臉部特徵如眼睛，卻可能會因為影子、眼鏡和人臉姿勢的關係導致偵測錯誤。而不同的臉部表情也會產生不同的臉部呈現樣式。這些因素增加了人臉偵測的困難性。本論文之主要目的在於提出一個解決上述問題的人臉偵測方法。 在本論文中，我們首先將會提出一個在多變化環境下的眼睛水平線偵測方法，利用膚色資訊將人臉可能所在的區域標記出來，然後根據人類眼睛擁有的非膚色特性與高亮度(gray-level)變化特性，將眼睛的位置標示出來，最後利用人臉的幾何特性將眼睛水平線標記定位。這個方法主要的貢獻在於提出一個有效的方法，取得多變化環境下的眼睛候選位置與水平線資訊。接下來，利用得到眼睛水平線位置與眼睛可能位置座標，發展一個在多變化環境下不同姿勢的人臉偵測方法。首先將膚色資訊所標記出來的人臉區域進行肩膀區域的判定與刪除，接著利用定義好的側面人臉特徵幾何規則將人臉區域分成側面人臉或非側面人臉區域。對於非側面的人臉候選區域，利用前述方法取得眼睛水平線與眼睛可能位置座標。最後針對不同人臉姿勢與人臉亮度分佈的特性，標記出真正的眼睛位置並藉此將人臉位置標示出來。本論文所提出的方法，經過標準人臉資料庫實驗證實，可以解決人臉偵測在多變化環境下所遭遇的問題。

While digital surveillance systems are receiving increasing concern in modern society, developing a fully automated face system is getting more and more attentions than before. However, a robust face detector is a foundation of building an automated face system. Face detection method insures a face system realizable. Thus, the demand for an efficient method to automatically detect face becomes urgent. Although face detection has been studied for more than 20 years, developing a human face detector under various environments is still a challenging task. An automatical face-detection job would include three steps: face candidate location, facial feature extraction and face detection. Some factors make face location difficult. One is the variety of colored lighting sources, another is that facial features such as eyes may be partially or wholly occluded by shadow generated by a bias lighting direction; others are races and different face poses with/without glasses. These factors make face detection difficult. The aim of the research is to provide a solution to these problems. In this thesis, we will firstly propose a method to extract the horizontal eye line of a face under various environments. Based on the facts that the eye color is very different from skin color and the gray level variance of an eye is high, some eye-like regions are located. Then the horizontal eye line is extracted based on the located eye-like regions and some geometric properties in a face. The contribution of this method is providing information of eye-like region positions and human horizontal eye line under various environments. Next, a method based on the extracted eye-like region position and horizontal eye line information will be proposed to detect a face with different poses under various environments. The basic idea is, first, skin regions are extracted from an input image using skin color information and then the shoulder part is determined and cut out by using shape information. The remained head part is identified as a face candidate. For a face candidate, we apply a set of geometric features to determine if it is a profile face. If the face candidate is a non-profile face, then a set of eye-like rectangles extracted from the face candidate and the lighting distribution are used to determine. Solving the poses problem and detecting face location under various environments are the main contributions of this method. Experimental results show that the proposed method is robust under a wide range of lighting conditions, different poses and races.