摘要
去除光照影响是人脸图像增强和识别的关键技术.二维经验模态分解(2-D EMD)可根据图像特征自适应地将其分成多个尺度的分量,各分量重建可实现图像增强或去噪的功能.基于二维EMD的原理及特征,研究了2种新的人脸图像去光照算法,即基于原始图像的直接二维EMD重建和基于Lambert光照模型的对数二维EMD重建,并运用特征脸识别对去光照算法的效果进行了测试和比较分析.仿真结果表明,运用二维EMD重建可较好地去除人脸图像中的光照影响,而对数二维EMD重建算法相对具有更明显的稳健性和适用性.
De-illumination is a key technology for enhancing and recognizing images of human faces. An image can be decomposed into several intrinsic mode functions （IMFs） adaptively by two dimensional empirical mode decomposition （2-D EMD）. The reconstruction of these IMFs is useful in image denoising and enhancement. Based on the theory and features of 2-D EMD, two new kinds of de-illumination methods were explored ： 2-D EMD followed by reconstruction of the original image, and 2-D EMD followed by reconstruction of a log-image based on the Lambert illumination model. A test and comparison using the eigen-face recognition method was run. The results showed good de-illumination performance by both methods, but log-images had better robustness and practicability.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2009年第12期1425-1429,共5页
Journal of Harbin Engineering University
基金
国家自然科学基金资助项目(60672034)
高等学校博士点基金资助项目(20060217021)
黑龙江省自然科学基金资助项目(zjg0606-01)
