【摘要】 针对人脸识别中的光照问题,本文提出一种多分辨率分层的快速处理算法。该算法建立在扩展的全变分模型的理论基础之上,具有模型简单,优化参数少等优点,并能有效的减少光晕现象。算法从快速性入手,首先将光照劣化图像进行多分辨率分层,在每一层上利用模型进行估计,从而得出整体的光照成分,然后恢复出人脸的反射系数图像,并以此作为光照不变量来进行识别。实验证明,通过多分辨率分层的方法,不仅能使系统在不同光照条件下的识别率得到提高,而且还能解决单独用TV(Total Variation)模型进行估计时处理速度较慢的缺点。更多还原

【Abstract】 Illumination is one of the most difficult problems in Face Recognition.In this paper,we present a multi-resolution algorithm for illumination processing,which is based on Total Variation Model.Compared to other illumination algorithms,TV model has some advantages:(1) It has good edge-preserving property,so as to remove "halo" efficiently.(2)There are only a few parameters to set.But there exists an obvious defect: because TV model operates on the whole image by itself,the process of estimating the lighting image will consume much time.Our algorithm adopts a multi-resolution strategy,greatly improving the speed.Experiments results show that our algorithm can achieve high recognition rate and speed simultaneously under various lighting conditions.更多还原