At present, the image super-resolution (SR) method based on sparse representation has the problem that the reconstruction speed and quality are difficult to be achieved simultaneously. Therefore, this paper proposes an improved anchor neighborhood regression SR algorithm based on low-rank constraint. Firstly, considering the critical role of locality in nonlinear data learning, the locally weighted regularization weight is introduced in the calculation of the projection matrix, which can constrain the projection process according to the correlation between the anchor point and the atoms in the corresponding neighborhood. Then, in the reconstruction phase, based on the assumption of low-rank between similar blocks, further constraints are made on the reconstruction blocks to obtain better reconstruction image quality. Experiments show that our method can not only reconstruct more image details but also achieve better reconstruction speed. Compared with some state-of-the-art sparse representation method, it achieves better reconstruction results in objective evaluation criteria.

当前，基于稀疏表示的图像超分辨率（SR）方法存在难以同时实现重建速度和质量的问题。因此，本文提出了一种基于低秩约束的改进的锚邻域回归SR算法。首先，考虑到局部性在非线性数据学习中的关键作用，在投影矩阵的计算中引入了局部加权正则化权重，可以根据锚点与相应邻域中原子之间的相关性来约束投影过程。然后，在重建阶段，基于相似块之间的低秩的假设，对重建块进行进一步的约束以获得更好的重建图像质量。实验表明，该方法不仅可以重构更多的图像细节，而且重构速度更快。与一些最新的稀疏表示方法相比，它在客观评估标准中获得了更好的重建结果。