局部自相似性与MRF网络模型的超分重建算法

针对局部自相似性重建方法的块效应问题,以及MRF网络模型方法外部训练库数据不相关性产生的图像重建误差问题,提出了一种结合局部自相似性和MRF网络模型的超分辨率重建方法。首先,利用图像局部自相似特性,引入自身冗余信息构建高分训练库,然后建立低分与高分训练库映射的MRF网络模型,通过置信传播算法求解MRF模型重建高分图像。以仿真和真实卫星图像进行超分实验,结果表明本文方法能够改善图像的细节,较好地去除了块效应,提高了地物边缘的清晰度。

Asuper-resolution algorithm based on local self-similarity and MRF model

Aiming at the block effect problem of local self-similarity reconstruction method and the image reconstruction error caused by the data irrelevance of the external training library of MRF network model method,a super-resolution reconstruction method combining local similarity and MRF network model is proposed Firstly,the image local self-similarity is used to construct the high-level training library,and then the MRF network model is established between low and high training library.The MRF model is solved by the belief propagation algorithm.The results show that this method can improve the detail of the image,remove the block effect and improve the clarity of the feature edge.