用神经网络恢复残差的图像超分辨率算法

Super-Resolution of Images Based on Restoration of Residual Errors by Neural Networks

本文提出了一种基于残差线性恢复的图像超分辨率算法框架 ,以已有的图像插值算法为基础 ,利用人工神经网络恢复图像残差 ,相迭加得到高分辨率图像的估计 .在数学上论证了基于神经网络的图像插值算法的依据 ,并证明了本算法比以往的算法具有更好的性能 .理论上 ,任何一种单帧的图像插值算法都可以被引入本算法框架 .同时给出了算法实例和测试结果 .

A super-resolution scheme frame based on linear restoration of residual errors is proposed, which combines an existing image interpolation algorithm with an artificial neural network (ANN) modeling the residual errors between the interpolated image and the real high-resolution image. Mathematical proof is presented that the performance of our method is better than existing interpolation algorithms. Theoretically, any intra-frame interpolation algorithm can be combined into our scheme to improve its performance. Example applications and their results demonstrate the effectiveness of our approach compared with traditional methods.