复杂背景下的人脸分割  

Face segmentation in complex backgrounds

作　　者：毋立芳[1] 沈兰荪[1] 孔潇[1] 朱旭娟[1] 

机构地区：[1]北京工业大学信号与信息处理研究室,北京100022

出　　处：《信号处理》2003年第2期100-103,共4页Journal of Signal Processing

基　　金：863项目(2001AA111391);国家自然科学基金(90104013)支持

摘　　要：针对复杂背景下人脸位置的不确定性和人脸轮廓边缘点的不连续性,本文提出一种新的人脸分割算法,能够在复杂背景下将人脸快速分割出来。本文算法包括三部分:首先设计了特征与模板相结合的人脸定位算法,在复杂背景中以双眼和嘴的坐标为基准确定人脸位置;其次,设计了自适应搜索算法,提取真正的人脸轮廓点,去除“假”轮廓点;最后利用人脸轮廓的平滑性通过曲线拟合来补充不确定轮廓点,完成人脸分割。本文算法综合利用了人脸的肤色、表面结构、轮廓边缘以及轮廓的平滑性等特征完成人脸分割,实验结果证明了算法的有效性。In this paper, a new face segmentation algorithm is put forward to deal with the uncertainty of face location and the discontinuity of face contour.The algorithm includes three parts: Firstly, a face location algorithm is put forward to locate faces in a complex backgrounds, where face location is represented by the centers of two eyes and mouth. Secondly, an adaptive searching scheme is designed to extract the true contour points of the face, and to discard false contour points. Finally, Based on the true contour points, polynomial curve fitting is made, moreover, based on the result of curve fitting, face region can be segmented accurately. In the algorithm, the human complexion, the structural feature of face, the edge feature of face contour and the smoothness of face contour are used, Experiment results confirm the efficiency of algorithm.

关 键 词：人脸分割 人脸定位 曲线拟合 自适应搜索算法 MPEG4 图像处理 人脸图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]