自適性閥值的人臉偵測系統

Abstract

本篇文章提出一個簡單且強健的方法去偵測人臉的位置，由於複雜的背景嚴重影響到人臉偵測的成果。我們首先使用膚色分割和連通成分的方法找出人臉候選區、再使用色彩分析的方法從人臉候選區中去尋找臉部特徵(眼睛、嘴唇)。使用色彩的資訊可以減少檢測時間，不過在不同的光線下會有誤判的情況產生，所以我們用自適性的閥值來找出臉部特徵，針對不同的影像用不同閥值去做判斷，最後利用眼睛和嘴唇的幾何關係找到正確的人臉位置。實驗結果除了在臉部特徵被遮蔽的情況下會有偵測不到的狀況，其他情況都有良好的偵測率。

We proposed a simple and robust method to detect the faces location of people, because the complex background influences the result of detection seriously. First we use the skin color segmentation and connected component method to find out the face candidate and then use color-analysis to search the features of face (eyes , lip) from face candidate. Using color information could save time of detection, but there will cause wrong detection while under different light conditions, so we use an adaptive threshold value to find out the face features, and using different threshold values to determine face features of different images, utilize the geometry relation between eyes and lips to locate correct face finally. The experimental result all have good detecting rate besides the face features be covered.