摘要
针对人脸识别中的特征提取问题,本文提出了一种由粗到精快速准确的嘴部自动检测和定位方法。该方法首先通过Adaboost算法检测出人脸图像大致的嘴部区域,缩小了后续定位的搜索范围;采用局部梯度算子提取嘴部轮廓,通过Ostu阈值法对提取的轮廓进行二值化处理,根据链码跟踪最终确定左右嘴角的精确位置。实验结果表明,该方法自动检测和定位嘴部快速准确,对表情和姿态的影响具有比较高的鲁棒性,有助于提高人脸识别算法的识别率。
A fast and accurate method of automatic detection and location of the mouth is proposed according to the feature extraction of face identification. At first, mouth region is roughly detected by Adaboost in face image, so the range of search is reduced. Then, mouth contour is detected by local gradient operator and binarized by Ostu threshold method. Finally, the mouth comers are accurately localized by Chain code tracking. The experiment results indicate that this method can detect and locate mouth comers fast and exactly. It can reduce the influence of facial expression and posture and improve the recognition rate of face identification.
出处
《光电工程》
CAS
CSCD
北大核心
2009年第10期135-140,共6页
Opto-Electronic Engineering
基金
高等学校博士学科点专项科研基金(20070701020)
关键词
人脸识别
特征提取
嘴角定位
局部梯度算子
face recognition
feature extraction
mouth comer location
local gradient operator
