基于稀疏表示的超分辨率图像重建

Image super-resolution via sparse representation

为提高现有图像的分辨率,在Yang等提出的基于稀疏表示的图像重建方法的基础上进行改进,提出联合高低分辨率块求取稀疏系数的方法。使用初次重建的高分辨率块和输入的低分辨率块在联合字典下再次稀疏分解,用联合分解出的新系数和高分辨率字典重建最终的高分辨率块,该方法可通过迭代使用提升重建性能。实验结果表明,改进的方法在目视效果和PSNR及SSIM二项指标值上都比Yang等方法更佳。

To improve the resolution of the existing images,basing on the reconstruction method proposed by Yang et al,which is based on sparse represent,an approach combining high and low resolution patches to calculate sparse represent coefficients was presented.The high resolution patches reconstructed initially and the input of low resolution patches were used together to calculate the sparse coefficients in the high-low-resolution-union dictionary,and the final high resolution patches were reconstructed using the high resolution dictionary and the coefficients.This method could be iteratively used for further performance improvement.Experimental results show the proposed method is better than Yang＇s method in both the visual effect and the PSNR and SSIM values.