近年來,人臉辨識在安全識別上及許多應用所提出之研究受到廣泛重視,而人臉偵測為人臉追蹤與識別的重要前置作業。雖然已經有許多的方法被用來實行人臉偵測的工作,但仍然存在著許多因素與環境條件使得人臉偵測的困難度居高不下,例如人臉的大小、傾斜方向、側臉問題、光源不足、是否重疊、臉部表情及臉部是否有裝飾物如眼鏡等等。本篇提出一個以膚色為基礎並且結合了臉部各種特徵與應用樣板比對的技術來做為人臉偵測的策略。由於本篇是針對在光源不足的情況下做偵測,因此一開始在膚色分析上進行分類,將分類上先以HSI色彩模型粗略的將影像分成正光與不正常光源,但受暗沉光源影響偵測過後之膚色部份並不完全,根據光源分類結果再做細微的修補以利之後的分析,中值濾波在此也加以利用,之後為了更進一步排除非人臉部份,人臉唇色偵測與排除與臉部三角幾何關係來得到進階的人臉可能的位置,最後再利用鼻子因光源產生的特殊梯度進行比對分析來將人臉區域偵測出來。 在實驗結果當中,針對光源不足之照片做為處理,其結果都能有效的克服因光線不足所造成複雜背景與膚色相近的問題,來得到準確的人臉區域。
In recent years, face recognition and tracking have been employed in many applications and research areas such as security identification, etc. Face detection is an important preprocessing task of them. In this paper, we propose a method based on the feature of skin color and other important information for detecting the correct position of human face under the situation of insufficient light source. In our method, images are first classified by the feature of skin color and then finely repaired for further analysis. Next we roughly detect the candidate position of the eyes and lip. After excluding the improper triangular relation of them, it is mapped into a regular triangle using affine transformation. And then we search the nose position using the pattern of vertical and horizontal gradient. At last the face area is determined by the eyes, nose and lip positions. The experiments show that our method is able to detect the position of human face effectively. Moreover, our method can overcome the problem that the colors of complex background are similar to the skin caused by the insufficient light source. Thus our method is then effective and efficient for face detection.