基于多引导滤波器的单幅图像超分辨率技术  

Single Image Super-Resolution Based on Multi-Guided Filtering

作　　者：刘哲[1,2] 韩九强 黄世奇 

机构地区：[1]西京学院电子信息工程系,陕西西安710123 [2]西安交通大学电信学院计算机科学与技术系,陕西西安710049

出　　处：《红外技术》2017年第10期920-927,共8页Infrared Technology

基　　金：国家自然科学基金(61473237)

摘　　要：提出了一种基于多引导滤波器的单幅图像超分辨率方法。首先,该方法通过大量的自然图像建立高低分辨率图像块样本训练库,并通过聚类算法将具有相似性质的高低分辨率样本块进行聚类;其次,将输入低分辨率图像进行重叠分块,并在样本库中搜索最近邻的高低分辨率样本聚类;再次,将输入低分辨率图像块作为输入图像,与样本库中最近邻的低分辨率聚类样本作为引导图像,运用本文提出的多引导滤波器计算引导滤波器的参数;最后,利用样本库中最近邻的高分辨率聚类样本和引导滤波器的参数,通过多引导滤波器就可以重构高分辨率图像。实验结果表明,本文算法不仅能很好地重构图像的高频细节,还能很好地恢复图像的纹理特征。In this paper, a single image super-resolution method based on multi-guided filtering is proposed First, an exemplar training database consisting of pairs of low-resolution and corresponding high-resolution image patches is constructed using many natural images. High- and low-resolution image patches with similar properties are clustered using a clustering algorithm. Next, the low-resolution input image is divided into overlapping patches, and the nearest neighbor high-and low-resolution sample cluster is searched against the exemplar training database. Then, the low-resolution input image patch is used as the new input; the nearest neighbor low-resolution clustering sample is used as the guide image. The mul- ti-guided filter is used to calculate the parameters of the guide filter. Finally, the high-resolution images can be reconstructed with the multi-guided filter using the nearest neighbor high-resolution clustering samples and the parameters of the multi-guide in the sample bank. Experimental results show that the proposed algorithm not only reconstructs the high-frequency detail of an image, but also recovers the texture features.

关 键 词：超分辨率 引导滤波器 样本训练库 高频细节 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]