基于小波光照归一化和高判别力特征的人眼定位算法  

Eye Location Algorithm Based on Wavelet Theory and High Discrimination Features

作　　者：高志升[1] 岳桢 张铖方 胡占强[1] 

机构地区：[1]西华大学无线电管理技术研究中心,四川成都610039

出　　处：《西华大学学报（自然科学版）》2015年第3期1-5,12,共6页Journal of Xihua University:Natural Science Edition

基　　金：国家自然科学基金项目(61372187;61473239);四川省教育厅重点项目(14ZA0118)

摘　　要：人眼定位是人脸识别、人脸分析等的前提,但是人眼定位的精度极易受到非均匀光照和噪声的影响。针对这一问题,提出一种结合小波变换光照归一化和高判别力特征的人眼定位算法。首先通过小波变换对人脸图像进行光照归一化处理,然后对人眼候选区域提取LTP和LPQ特征,对每个候选人眼区域计算SVR分类响应值得到人眼概率图,最后对人眼概率图进行高斯拟合,完成人眼精确定位。在人脸数据集CMU-PIE、Yale B和AR上的实验结果表明,该算法能有效克服非均匀光照对人眼进行精确定位的影响,具有光照变化鲁棒性,比同类算法有更高的定位精度。Eye location is the prerequisite for face recognition and face analysis. However, the accuracy of eye location is vulnera- bly affected by non-uniform illumination changes and noise. To solve this problem, this paper proposes an algorithm based on wavelet theory and high discrimination features for eye location. Firstly, face images undergo illumination normalization utilizing wavelet theo- ry. Secondly, the LTP and LPQ features are extracted from the eye-candidate regions, and then the eye probability map is subsequently obtained from the calculation of the responding classification value of SVR for each eye-candidate point. Lastly, applied Gaussian fitting method to the eye probability map and the human eye is located accurately. Extensive experiments on CUM PIE, Yale B and AR Face Databases demonstrate that our method can effectively overcome the bad effects of non-uniform illumination changes and noise for the accurate eye location, and improve the robustness to changes in illumination and outperform the state-of-the-art approaches in terms of the accuracy of eye localization.

关 键 词：小波 LTP LPQ SVR 高斯拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]