基于PDE正则化的超分辨率图像重构方法

Image Super-resolution Using PDE Regularization

采用非线性扩散模型建立超分辨率图像重构的偏微分方程,利用各向异性扩散方程的方向选择平滑的特性,在重构高分辨率图像的同时能够很好地消除系统噪声,保持细节信息。实验结果表明,该方法有效地提高了重构的图像质量,在视觉观察和数值评价上都优于原有正则化方法,并且对不同噪声水平的图像具有很好的鲁棒性。

From the typical anisotropic diffusion equations, the PDEs for image super-resolution are deduced and, then a robust regularization-type reconstruction algorithm is devised. The proposed approach can suppress the system noise and preserve more details of the estimated HR image by orientation-selective regularization. The experimental results show that the PDE-based approach makes a promising progress in image super-resolution. It outperforms the previous methods in both numerical and perceived visual quality and is robust to noisy images.