【摘要】 对于模糊造成的严重退化的图像，重建出令人满意的高分辨率图像仍然是一个挑战。直观的解决方案是引入图像预处理模块，提出了一个基于约束最小二乘方滤波的图像预处理模块。该模块通过估计的模糊核和自适应滤波参数生成去模糊特征。预处理后的干净特征和LR图像先验特征输入特征融合网络，恢复最终的高分辨率结果。在合成退化图像的定量和定性上都优于其他主流方法，在真实世界图像的实验中，所提的算法在视觉效果上也更具优势。更多还原

【Abstract】 Reconstructing satisfactory high-resolution images for severely degraded images caused by blurring remains a challenge. An intuitive solution is to introduce an image preprocessing module, which is proposed based on constrained least squares filtering. This module generates deblurred features by means of estimated blur kernels and adaptive filtering parameters. The preprocessed clean features and LR image prior features are input to a feature fusion network to recover the final high-resolution results.The proposed algorithm outperforms other mainstream methods both quantitatively and qualitatively in synthesizing degraded images, and it is also more advantageous in terms of visual effects in experiments with real-world images.更多还原