中文摘要：

针对目前基于稀疏表示的常用图像融合算法计算复杂度高以及忽略图像局部特征的问题,提出多尺度稀疏表示（multi-scale sparse representation,MSR）的图像融合方法。充分利用小波多尺度分析较好突出图像局部特征的特点,将其和过完备稀疏表示有效结合;待融合图像在小波解析域中进行小波多层分解,对每个尺度的特征运用K-SVD（kernel singular value decomposition）多尺度字典进行OMP（orthogonal matching pursuit）稀疏编码,并在小波域中各个尺度中进行融合。实验结果表明,与传统的小波变换、轮廓波变换、稀疏表示融合算法相比,该算法更能保证图像局部特征的完整性,实现更好的性能。

英文摘要：

The fusion algorithm based on the sparse representation is highly computational complex and ignores the local charac- teristics of the image. Hence a method based on the multi-scale sparse representation was proposed to overcome these disadvanta- ges. The wavelet transform was combined with the sparse representation effectively, so that the local characteristics were studied better through the wavelet multi-scale analysis. The preparing image fusion was decomposed using wavelet multi-level decompo- sition in the wavelet analytic domain. Then the sparse coding for characteristics of each scale was represented by using multi-scale K-SVD dictionary with OMP, and each scale of image was fused in the wavelet domain. The experimental results show that this algorithm can ensure the integrity of the local characteristics of image and achieve better performance compared with the tradi- tional wavelet transform, contourlet transform and sparse representation.