中文摘要：

近年来，基于样本的图像超分辨率重建逐渐成为研究热点，该算法一般利用外部训练样本，测试图像与训练样本的相似度在一定程度上影响着重建结果。针对此类问题，提出一种基于局部回归和自相似性的图像超分辨率重建算法。应用不同尺度图像间的自相似特性，对图像块建立一阶回归模型完成重建的算法，充分利用图像自身信息，并用稀疏表示的方法替代遍历搜索自相似块的方法，可以在自相似块不足的情况下保证重建质量。实验结果表明，该算法的重建质量较高，可以一定程度减少外部训练样本带来的虚假高频问题，且在重建质量与重建时间上有着较好的折中。

英文摘要：

In recent years,image super-resolution reconstruction based on samples has gradually become a hot research topic, which usual- ly uses the external training samples. The similarity between the test image and the training samples affects the reconstruction results to a certain extent. To solve this problem,a super-resolution image reconstruction algorithm based on local regression and self-similarity is proposed. This algorithm, which makes use of the self-similarity between images at different scales and reconstructs the image by establishing the first-order autoregressive model of the patches,could make full use of the information of the image itself, and replace the traversal search of self-similar patches with the sparse representation method. So it can guarantee the reconstruction quality even the number of the self-similar patches is not enough. The experimental results show that the reconstruction quality of this algorithm is high. It can alleviate the false high-frequency problem brought by the external training samples to a certain extent and have a good tradcoff between the reconstruction quality and reconstruction time.