光学学报, 2017, 37 (3): 0318006, 网络出版: 2017-03-08
自外而内的单幅图像超分辨率复原算法 下载: 976次
Single Image Super-Resolution Restoration Algorithm from External Example to Internal Self-Similarity
图像处理 超分辨率 样例学习 深度卷积网络 自相似 image processing super resolution example-based methods convolutional neural network self-similarity
摘要
单幅图像超分辨率(SR)复原是一个病态逆问题, 需要利用图像的先验知识进行正则化约束。提出了一种同时考虑外在样例和内在自相似性的单幅图像SR复原算法, 其中外在先验知识是通过卷积神经网络从外在低分辨率-高分辨率图像对学习得到的, 而内在先验约束由聚类和低秩近似实现。实验结果表明, 本方法在复原效果和稳健性方面优于已有方法。
Abstract
Single image super-resolution (SR) restoration is an ill-posed inverse problem, in which regularization restriction is done with image priori knowledge. One single image SR method is proposed which simultaneously taking external example and internal self-similarity into account. Here the external knowledge is learned by convolutional neural network from external low-resolution-high-resolution image pairs, while the internal prior is utilized by cluster and low-rank approximation. The experimental results show that the proposed method outperforms many other existing super-resolution methods in recovery effect and robustness.
郑向涛, 袁媛, 卢孝强. 自外而内的单幅图像超分辨率复原算法[J]. 光学学报, 2017, 37(3): 0318006. Zheng Xiangtao, Yuan Yuan, Lu Xiaoqiang. Single Image Super-Resolution Restoration Algorithm from External Example to Internal Self-Similarity[J]. Acta Optica Sinica, 2017, 37(3): 0318006.