摘要
光照变化对单样本人脸识别的效果影响巨大。对于光照问题,Weber-Faces的局部光照相同的假设在光照突变下不是非常严谨。为了消除光照影响,提高人脸识别率,针对这一假设问题,提出一种基于γ变换的Weber-Faces光照归一化方法。通过γ变换有效改善人脸图像的光照条件,达到局部光照均匀化的目的,使得局部光照相同的假设更加合理;通过Weber-Faces的方法进行光照归一化,能够消除光照分量,得到对光照更加鲁棒的反射人脸;提取反射人脸的LBP特征并进行识别。在光照角度变化的Extend Yale B人脸库和光照强度变化的AR人脸库上进行实验。实验结果证明,该方法对光照具有更好的鲁棒性,能够提高Weber-Faces的人脸识别率。在两个数据库上的识别率可以分别达到99. 04%和99. 56%,高于Weber-Faces的98. 94%和99. 50%,并优于其他光照归一化方法,如SSR、GIC、SQI、MSW等。
Illumination changes have great influences on the recognition effect for single sample face recognition. For illumination problems, the assumption proposed by Weber-Faces that the local illuminations are the same is not rigorous on illuminations mutation. Aiming at the assumption, we proposed Weber-Faces illumination normalization method based on γ transform to eliminate the influences of illumination and improve the face recognition rate. In the method, the illumination conditions of the face images were improved by the γ transform, and the local illuminations were homogenized, making the assumption that the local illuminations were the same more reasonable. The illuminations were normalized by the Weber-Faces method, and the illumination component could be eliminated, and the reflection face which was more robust to the illumination was obtained. The LBP features of the reflection face were extracted for recognition. Experiments were performed on the Extend Yale B face database with changing illumination angle and AR face database with varying illumination intensity. The experiments show that the method is more robust to lighting, and can improve the face recognition rates of Weber-Faces. The recognition rates on two databases can reach 99.04% and 99.56% respectively. It is higher than 98.94% and 99.50% of Weber-Faces, and is superior to other illumination normalization methods, such as SSR, GIC, SQI and MSW.
作者
杨超
伍世虔
方红萍
Yang Chao;Wu Shiqian;Fang Hongping(School of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430081,Hubei,China;Hubei Collaborative Innovation Center for Advanced Steels,Wuhan 430081,Hubei,China;School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,Hubei,China)
出处
《计算机应用与软件》
北大核心
2018年第12期174-178,共5页
Computer Applications and Software
基金
国家自然科学基金项目(61371190
61775172)
关键词
光照归一化
γ
变换
韦伯人脸
人脸识别
特征提取
Illumination normalization
γ transform
Weber-Faces
Face recognition
Feature extraction
