基于扩散的自适应超分辨率重建

Adaptive supper-resolution reconstruction based on diffusion

从各向异性角度分析了P-M模型、L1范式(TV模型)、L2范式(调和模型)的不足,通过扩散模型建立超分辨率重建的偏微分方程,提出一种非线性各向异性和超分辨率重建组合的模型。该模型在图像平坦区域具有线性各向同性扩散,能够有效消除噪声,在图像边缘区域具有非线性各向异性扩散保留边缘,有效减少了滤波产生的阶梯效应和P-M模型过渡平滑忽略细节的现象。仿真结果表明,该模型能够有效地提高图像重建质量,能在消除噪声的同时保留边缘,具有很好的鲁棒性。

The disadvantages of P-M model,L1norm(TV model)and L2norm(harmonic model)are analyzed in the as-pect of anisotropy.The partial differential equation for super-resolution reconstruction is established with diffusion model.A modelcombining the nonlinear anisotropy and super-resolution reconstruction is proposed.The model has the characteristics of linearisotropic diffusion in the flat area of the image,which can eliminate the noise effectively,and has nonlinear anisotropic diffu-sion preserving edge in the edge area of the image,which can effectively reduce the staircase effect produced by filtering andavoid the phenomenon that the details of P-M model is neglected due to transition smooth.The simulation results show that themodel can improve the image reconstruction quality effectively,eliminate the noise while remaining the edge,and has good ro-bustness.