NSCT域自适应人脸图像光照不变特征提取  

Adaptive extraction of illumination invariant from face images in NSCT domain

作　　者：赵振华[1,2] 郝晓弘[1,2] 

机构地区：[1]兰州理工大学电信学院,甘肃兰州730050 [2]兰州理工大学甘肃工业先进过程控制重点实验室,甘肃兰州730050

出　　处：《光电子．激光》2012年第9期1815-1820,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61064003)资助项目

摘　　要：为了减少光照变化对人脸识别算法的影响,提出了一种基于非下采样Contourlet变换(NSCT,nonsubsampled contourlet transform)的光照不变特征提取方法。人脸图像经过对数变换(LT)后,利用NSCT进行分解,得到图像的低频子带和高频方向子带;根据高频子带中NSCT系数的概率分布,给出各子带的自适应阈值,并采用折衷阈值函数进行滤波;对滤波后的子带进行NSCT逆变换,得到人脸图像的光照不变特征。在Extended Yale B和CMU PIE人脸数据库上的实验结果表明,本文方法能有效减少光照影响,显著提高了识别率。In order to alleviate the influence of illumination variations on face recognition,a novel method is proposed for extracting illumination invariant from face images. In this work,the logarithm of an original face image is first decomposed by using nonsubsampled contourlet transform （NSCT）, which is a fully shift-invafiant,multi-scale and multi-direction transform. Then each high-pa^s sub-band in NSCT domain is filtered by adaptive thresholds designed according to the probability distribution of the coeffi- cients in the sub-band. Finally, the illumination invariant is obtained by using reverse NSCT. Experimental results on the extended Yale B and the CMU PIE face data sets show that the proposed method achieves satisfactory recognition rates under varying illumination conditions.

关 键 词：光照不变量 自适应阈值 人脸识别 非下采样Contourlet变换(NSCT) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]