一种光照鲁棒性人脸图像超分辨率算法  

A Robust Face Image Super-resolution Algorithm for Illumination

作　　者：马祥[1] 马琴琴 付俊妮 MA Xiang;MA Qinqin;FU Junni(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《计算机工程》2018年第9期199-202,共4页Computer Engineering

基　　金：国家自然科学基金(61771075)

摘　　要：为解决光照变化人脸图像的超分辨率问题,提出一种图像超分辨率算法。将输入的低分辨率人脸图像和人脸图像训练集相结合,在低分辨率空间通过对角加载冗余转换,产生多种不同光照的低分辨率人脸图像,并进行局部几何位置约束重建,加权合成多种不同光照的高分辨率人脸图像。实验结果表明,在将人脸图像分辨率提高4×4倍的情况下,该算法不仅能在低分辨率空间下,对一幅分辨率极低的人脸图像重建出所对应的多个不同光照的低分辨率人脸图像,而且能够重建出多种不同光照下质量较高的高分辨率人脸图像。In order to solve the problem of face image super-resolution when the illumination changes,a robust face super-resolution algorithm for illumination is proposed.Combining the input low-resolution face image with the face image training set,the low resolution face images under multiple illuminations are generated by the diagonal loading redundant transformation in the low resolution space,and the low resolution face images under multiple illuminations is reconstructed by geometry and position constraints.When the optimal weight is solved,the high resolution face images under multiple illuminations are synthesized.Experimental results show that the proposed method cannot only produce the low resolution face images with multiple illuminations for a very low-resolution face images in low resolution space,but generate satisfying high-resolution face images of multiple illuminations.

关 键 词：光照变化 超分辨率 对角加载 冗余转换 几何位置约束 

分 类 号：TP911.73[自动化与计算机技术]