摘要
将小波域多尺度分析的思想和图像域单尺度稀疏表示的思想有效结合,提出基于多尺度字典学习的图像融合方法。首先将训练图像变换到小波域,分别对各个子带系数训练字典;根据训练的字典求解并融合源图像各个子带的稀疏表示系数;经过逆小波变换重构融合图像。提出的方法综合了学习字典的稀疏特性和小波分析的多分辨率特性。实验结果表明较现有基于图像域字典学习的融合方法和基于小波域多尺度分析的融合方法均具有更优的融合效果。
We combine the multi-scale analysis in wavelet domain with the single-scale sparse representation in image domain and propose an image fusion algorithm based on multi-scale dictionary learning. We transform the trained images into wavelet domain and train the dictionary for each sub-band dictionary. We use the trained dic- tionary to solve and fuse the sparse representation coefficient of each sub-band of a source image. The fused image is reconstructed through the inverse wavelet domain. Our algorithm combines the sparse character of a learned dictionary with the multi-resolution character of wavelet analysis. The experimental results, given in Fig. 2 and Table 1, and their analysis show that our image fusion algorithm outperforms those based on the learned dictionary in image domain and multi-scale analysis in wavelet domain respectively.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2013年第5期793-797,共5页
Journal of Northwestern Polytechnical University
基金
国家自然科学基金(61075014)
航天支撑基金(2011XW080001C080001)
西北工业大学博士论文创新基金(CX201318)资助
关键词
图像融合
多尺度字典学习
稀疏表示
K—SVD
algorithms, image fusion, image processing, wavelet transforms
image domain, muli-scale analysis, muli-scale dictionary learning, sparse representation, wavelet domain
