Multi-frame image super-resolution reconstruction based on spatial information weighted fields of experts
Multi-frame image super-resolution reconstruction based on spatial information weighted fields of experts
复制标题
基于专家空间信息加权场的多帧图像超分辨率重建
DOI：
10.1007/s11045-019-00648-5
复制
发表时间：
2020-01-01
影响因子：
2.5
通讯作者：
Lin, Pan
中科院分区：
文献类型：
--
作者：
Huang, Shuying;Wu, Jiajun;Lin, Pan
针对传统的专家领域（FoE）模型在去噪过程中会模糊图像边缘和纹理的局限性，提出了一种空间信息加权FoE（WFoE）模型，将图像的空间结构信息引入FoE模型。在边缘和平滑区域，根据曲率差构造单调递减函数来控制滤波器的权值。所提出的WFoE模型可以更好地去除噪声，同时保持边缘。此外，所提出的WFoE模型被设计为基于最大后验的多帧图像超分辨率（SR）重建算法中的正则化项，从而能够开发新的SR方法。由于WFoE模型更倾向于保留图像边缘，因此所提出的基于WFoE的SR重建方法在保留图像边缘方面比传统FoE模型可以获得更好的结果。实验结果表明，该方法具有更好的峰值信噪比和视觉逼真度相比，现有的一些SR方法。
To overcome the limitations of the traditional fields of experts (FoE) model, which will blur image edges and texture during the denoising processing, a spatial information weighted FoE (WFoE) model has been presented to introduce the image spatial structure information into the FoE model. A monotone decreasing function is based on the curvature difference to control the filter weight in the edge and smooth region. The proposed WFoE model can better remove noise while preserving edges. Additionally, the proposed WFoE model is designed as a regularization term in the maximum a posteriori-based multi-frame image super-resolution (SR) reconstruction algorithm, enabling the development of a new SR method. Since the WFoE model is more inclined to keep image edges, the proposed WFoE-based SR reconstruction method can obtain better results than traditional FoE model with respect to preserving image edges. Experimental results demonstrate that our method has better peak signal-to-noise ratio and visual verisimilitude compared with some existing SR methods.