基于二维经验模态分解的单幅图像超分辨率重建  

Single Image Super-resolution Reconstruction Based on Bidimensional Empirical Mode Decomposition

作　　者：毛晓波[1] 张志超[1] 

机构地区：[1]郑州大学电气工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2014年第5期15-18,共4页Journal of Zhengzhou University（Engineering Science）

基　　金：教育部高等学校博士学科点专项科研基金资助项目(20114101110005);河南省重大科技攻关计划资助项目(102101210100);河南省教育厅科学技术研究重点项目资助项目(14A410001)

摘　　要：针对已有的单幅图像超分辨率重建算法大都无法同时兼顾重建质量和运算速度的问题,提出了基于二维经验模态分解的单幅图像超分辨率重建算法.首先用二维经验模态分解法将一幅低分辨率图像分解为不同复杂程度的图层;然后对包含高频细节信息的第一个图层用改进核岭回归法重建,以保证重建质量;对包含较少信息的后几个图层用双三次插值法重建,以提高重建速度;最后用二维经验模态分解逆变换将重建后的各层图像合成一幅完整的高分辨率图像.实验结果表明该算法充分结合了三者的优势,在保证重建图像质量的同时,提高了算法的运算速度.Aiming at the problem that most learning-based super-resolution algorithms are not competent to take both the reconstruction quality and computing speed into account,a novel single image super-resolution reconstruction based on Bidimensional Empirical Mode Decomposition（BEMD） is proposed.Firstly,a lowresolution image is decomposed into several layers with different complexities by BEMD,which contains various details.Then,the first IMF be reconstructed with high frequency details by improved kernel ridge regression algorithm in order to ensure the quality of image and the other IMFs containing fewer details be reconstruct by bicubic interpolation in order to improve the speed.Finally,the reconstructed IMFs are merged into a highresolution image.Experimental results demonstrate that the proposed method make full use of the advantages of the above algorithms,which not only ensure the quality of the reconstruction image,but also improves the computational speed.

关 键 词：超分辨率重建 二维经验模态分解 改进核岭回归 双三次插值 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]