一种新的人脸归一化算法  

New algorithm for normalization of the face image

作　　者：靳薇[1,2] 张建奇[1] 梁建霞[2] 王加强[2] 

机构地区：[1]西安电子科技大学技术物理学院,陕西西安710071 [2]北京市新技术应用研究所,北京100021

出　　处：《西安电子科技大学学报》2011年第2期61-65,111,共6页Journal of Xidian University

基　　金：国家自然科学基金资助项目(60777042);北京市科学技术研究院萌芽计划资助项目

摘　　要：提出了一种基于改进的局部梯度算子进行嘴角定位、同时结合双眼位置进行人脸归一化的新方法.该方法首先通过Adaboost算法检测出人脸图像、眼睛和嘴部区域大致位置,然后采用灰度积分投影法精确定位两眼位置,并根据改进的局部梯度算子提取嘴部轮廓,最终依据两眼与嘴角的精确位置对人脸进行旋转和双向尺寸缩放处理.实验结果表明,该算法对特征点定位准确,能够更加精确地对人脸进行归一化处理.This paper presents a novel face normalization method by integrating the location of eyes and the corner of the mouth for the normalization in the vertical direction.In the first step,the regions of eyes and the mouth are roughly detected by making use of Adaboost all through the whole face image.The Grey Integral Projection and the improved Local Gradient Operator algorithms are then applied to find the accurate positions of the eyes and the corner of the mouth in the face image,respectively.Finally,the face image is rotated and zoomed according to the relative position of eyes and the mouth in vertical and horizontal directions.Experimental results demonstrate that the proposed method can locate the feature points accurately and present better normalization performance by introducing the zoom criterion in the vertical direction.

关 键 词：归一化 人脸识别 嘴角定位 局部梯度算子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]