经验模式分解及其应用于人脸图像光照预处理的几个关键问题研究

项目名称： 经验模式分解及其应用于人脸图像光照预处理的几个关键问题研究

项目编号： No.61202346

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 陈恒鑫

作者单位： 重庆大学

项目金额： 23万元

中文摘要： 光照变化是影响人脸识别效果的重要因素，特别是侧面光照在人脸图片上留下的阴影区域会严重干扰人脸本质特征的提取。经验模式分解（Empirical Mode Decomposition, EMD）可以将非平稳信号分解为不同频率成分的多个平稳信号之和，二维经验模式分解（Bi-dimensioin, BEMD）是EMD在二维信号上的扩展。人脸图像中的局部高频细节信息代表了人脸本质特征，而局部低频信息代表了光照信号，所以可以使用BEMD来分解人脸图像中不同频率分量，以达到进行光照预处理的目的。但是缺乏针对分解结果的量化评价标准、缺乏局部极值点、信号分量的迭代次数不确定，以及局部阴影造成的严重模式混淆现象等问题都会影响将BEMD用于人脸图像光照预处理的效果，本项目拟针对如何解决这四个方面的问题进行深入研究。

中文关键词： 经验模式分解；二维经验模式分解；去光照；人种分类；目标检测

英文摘要： Illumination variation is an important factor influencing the effect of face recognition. Especially, it is difficult work to extract the essential characteristic from the face image, in which there are some shadow regions. Empirical Mode Decomposition (EMD) can decompose non-stationary signal into several stationary signals which represent different frequency signal component. Bi-dimension EMD (BEMD) is an extensional version for 2D signal. In local area of face image, high frequency detail information represents the essential characteristic of face, and the low frequency information represents the illumination signal. So, BEMD can be used for illumination preprocessing, with the face image being decomposed into some frequency components. However, there are 4 important problems can decline that application of BEMD, and these problems are lacking evaluation criteria for decomposition, lacking local extreme point, uncertainty of sifting number for signal component and serious mode mixing caused by local shadow. This project will do much deep study for solving these 4 problems. Meanwhile, with the mode mixing problem being solved, the disturbance of face shadow will be suppressed.

英文关键词： EMD；BEDM；Eliminate Illumination Variant；Race Classification；Object Detection