一种基于EMD的人脸图像光照问题处理方法  

EMD-based illumination processing approach in face recognition

作　　者：王依才[1,2] 侯德文[1,2] 邱丽君[1,2] 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014 [2]山东省分布式计算机软件新技术重点实验室,济南250000

出　　处：《计算机工程与应用》2010年第27期166-168,共3页Computer Engineering and Applications

摘　　要：鉴于经验模态分解(EMD)算法的无参数特性和基于数据驱动的自适应分析,从EMD算法筛选过程中插值点和差值算法选取的角度出发,提出了一种标准EMD的改进算法。用改进的EMD算法分解需要识别的人脸图像,得到从整体上有效消除了光照变化影响的原人脸图像的重构图像,并将重构图像用于识别。通过在YaleB人脸库上的实验结果表明,将提出的光照处理方法应用到人脸识别中可以取得较好的效果。Empirical Mode Decomposition（EMD）,a fully data-driven method without using any predetermined filter or wavelet function, is a powerful tool for analyzing composite, nonlinear and non-stationary signals.Based on this, a method is described to address illumination effects in the biometric modality of face recognition using EMD to decompose images into their Intrinsic Mode Function（IMF）.Using these illumination modes the facial image is reconstructed without these illumination distortion components to synthesize a more illumination neutral facial image.Experimental results show that EMD is suitable for feature extraction and the proposed method demonstrates to be feasible for face recognition.

关 键 词：人脸识别 光照处理 经验模态分解 固有模式函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]