基于非局部低秩矩阵重建的图像插值

传统图像插值方法往往只考虑了图像局部相邻像素之间的关系进行插值，而忽略了图像中广泛存在的非局部自相似性。为了充分利用图像中的这种非局部自相似性以提高插值图像质量，本文提出了基于图像非局部低秩重建模型的图像插值方法,为低秩重建模型提出了一种基于分解为子问题交替迭代求解的高效求解算法。提出的算法能获得更高的主观与客观重建图像质量,实验表明,相对于Bicubic、SAI等图像插值方法能取得平均1.37dB和0.77dB的PSNR增益。

Image interpolation based on non-local low-rank matrix reconstruction

Traditional image interpolation method always only focuses on relationship between adjacent pixels of partial image to interpolate, ignoring non-local self-similarity extensively existing image. In order to fully utilize the non-partial self-similarity in image to improve quality of interpolated image, we propose a novel image interpolation method based on non-partial low-rank matrix reconstruction model of image, also present efficient iteration solution algorithm for low-rank reconstruction model. By the given algorithm, higher objective and subjective reconstructed image quality can be obtained. Furthermore it is shown from experiment that compared with Bicubic,SAI,etc., image interpolation method can generate PSNR gain with average of 1.37dB and 0.77dB.