基于多尺度自相似性和非局部均值的单幅图像超分辨率方法

Single Image Super Resolution Method Based on Multi-scale Self-similarity and Non Local Means

提出了一种基于多尺度局部自相似性和非局部均值的单幅图像超分辨率算法,该算法不依赖于外界图像,仅仅在原始图像的局部子窗口中搜索目标图像块的相似子块,利用非局部均值算法对相似子块进行加权求和来估计待复原图像,然后在复原图像上叠加最相似子块的高频细节图像,获得高分辨率图像。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 new single image super resolution algorithm is proposed based on multi-scale local self-similarity and non local means. This algorithm does not rely on an external example database nor use the whole input image as a source for example patches. Instead, similar patches are extracted from extremely localized region in the input image, similar patches are weighted summed to estimate the image to be restored using a non-local mean algorithm, and then high frequency detail image of the most similar patch is added to the restored image to obtain a high-resolution image. Experimental results show that the proposed algorithm can not only reconstruct the high frequency details of the image, but also restore the texture features of the image.