运用Foveated非局部均值和局部核回归的单幅图像超分辨重构

Single Image Super-Resolution Reconstruction with Foveated Non-Local Means and Local Kernel Regression

旨在对一幅低分辨图像进行单幅图像超分辨重构.采用Foveated距离度量图像的冗余相似块,利用Foveated非局部滤波进行Foveated非局部先验"冲突.利用Foveated非局部滤波进行Foveated非局部先验及引导核回归的局部先验,得出运用Foveated非局部均值和局部核回归的单幅图像超分辨重构方法.数值实验结果表明:该方法有效重构出清晰度较优的超分辨率图像,其图像边界、峰值信噪比及结构相似性均显著优于其他超分辨重构方法.

Both Foveated non-local and local regularization priors are learned to realize super-resolution reconstruction from a given low-resolution image. The patch similarities between redundant similar patches in natural images are measured through a Foveated distance. Foveated non-local prior is learned by using a Foveated non-local filter while the local prior is achieved by the kernel regression. A single image super-resolution reconstruction method is obtained with the use of Foveated non-local means and local regression. The experimental results show that the proposed method can effectively reconstruct better super-resolution images whose clarities are superior. Their image boundary,PSNR and SSIM are significantly better than those obtained by other super-resolution reconstruction methods.