基于相关滤波器组的人眼定位方法  

Eye Localization Based on Correlation Filter Bank

作　　者：葛仕明[1] 杨睿[1] 文辉[1] 孙利民[1] 

机构地区：[1]中国科学院信息工程研究所信息安全国家重点实验室,北京100093

出　　处：《网络新媒体技术》2014年第1期8-12,共5页Network New Media Technology

基　　金：中国科学院战略性先导项目(编号:XDA06040100);中国科学院信息工程研究所前瞻部署项目支持

摘　　要：人眼定位是人脸识别和认证中一个关键步骤,本文提出一种基于自适应聚类相关滤波组的人眼定位方法,将人眼定位表示成一个优化问题,并定义一个多滤波器的损失函数。多个滤波器采用自适应聚类方法训练来构造,得到的多个滤波器组成相关滤波器组。滤波器组包含多个模板,对人脸的差异具有良好的区分性,从而能够有效地定位变化较大时的人眼位置。与基于分类器的人眼定位方法以及传统的基于单一相关滤波器的方法相比,实验表明本文方法在检测准确率和定位精度上有显著提升。Eye Localization is a key step in many face verification applications. In this paper,we proposed a novel eye localization method based on trained correlation filter bank. We pose the eye localization as an optimization problem and define a correlation filter bank-based cost function. An adaptive clustering approach is used to train and construct the correlation filter bank. The trained filter bank includes several filter templates and is good at differential face variations,thus can provide accurate eye localization ability for variable poses,appearances and illuminations. Experiments compared with detector-based methods and traditional correlation filter-based methods show the advantages of the proposed method both in detection ratio and localization accuracy.

关 键 词：人眼定位 相关滤波 滤波器组 生物特征认证 人脸识别 聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]