基于核典型相关分析的图像放大算法

Image super-resolution via kernel canonical correlation analysis

为了更有效地映射高、低分辨率图像到变换域,提高图像的分辨率,给出一种基于核典型相关分析的单幅图像超分辨率放大算法。首先将训练高、低分辨率图像块矩阵映射到变换域上,利用核典型相关分析得到最优的变换域基向量,然后在该变换域上重建测试低分辨率图像,再转换到原空间上得到初始放大结果,最后利用迭代反投影算法进一步提高图像的质量。实验结果表明,新算法可提高图像的分辨率,并能重建出图像较多的细节,人工痕迹少。

In order to improve image resolution by mapping effectively high-and low-resolution image into transform domain,a single image super-resolution algorithm via kernel canonical correlation analysis is proposed.The algorithm maps the matrices of high-and low-resolution training image patches into transform domain and obtains the optimal base vectors via kernel canonical correlation analysis.Furthermore,the test low-resolution image is reconstructed in the transform domain and converted to the original space to gain the initial result.At last,an iterative back projection algorithm is used to further improve the image quality.Experimental results show that the new algorithm can improve the image resolution and reconstruct richer details and fewer artifacts.