基于肤色与结构特征的人脸检测与跟踪  

Face detection and tracking based on skin color and structure feature

作　　者：谢志鹏[1] 陈锻生[1] 

机构地区：[1]华侨大学计算机科学系,福建泉州362021

出　　处：《计算机工程与设计》2005年第11期3135-3137,共3页Computer Engineering and Design

摘　　要：首先使用基于YCbCr色彩空间的肤色高斯概率模型,对彩色图像进行肤色分割,得到候选人脸区域,接着提取候选区域的几何结构特征形成特征向量,利用bp神经网络进行候选人脸的判决,以此获得人脸的初始跟踪位置;后续的跟踪使用帧间差分法来确定人脸的移动方向和位置。该方法提高了人脸的检测速度和正确率。Using skin color Gaussin probability model based on YcbCr color space, the candidate face region is firstly segmented from the color image, then the geometric structual features was extracted from these regions and forming the respective feature vectors After that these vectors were input to BP neuron network to verify the true or false human face, therefore the initial face position was acquired. The succeeding face tracking adopted the frame differential method to figure out the direction and position of moving face. This method improves correctness ratio of detecting face and tracking speed.

关 键 词：肤色模型 高斯分布 特征提取 BP神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]