融合低秩和稀疏表示的图像超分辨率重建算法
Super Resolution Reconstruction Algorithm Combined Low Rank with Sparse Representation

针对现有图像超分辨率重建算法收敛速度慢、易受噪声影响的问题,结合低秩矩阵恢复与稀疏重建理论,提出了一种新的单幅图像超分辨率重建算法。对于待重建的退化图像,首先进行低秩恢复,得到含有原始图像大部分信息的低秩部分和主要由噪声组成的稀疏部分,然后对低秩部分利用学习的高低分辨率字典对进行稀疏重建。实验结果表明:本文算法对噪声鲁棒,运行速度快,图像视觉效果更佳;相比基于稀疏表示的统计预测模型(SPBSR),本文算法的峰值信噪比指标平均提高了4 dB。
Traditional super？？resolution (SR) reconstruction algorithm converges slowly, and is easily vulnerable to noise. We propose a novel single image SR reconstruction algorithm, which combines low rank matrix recovery with the sparse reconstruction theory. For a degraded image, the low rank part and the sparse part are obtained by the low rank matrix recovery theory firstly. The low rank part contains almost all information of the original image, and the sparse part is composed of noise. Then the sparse reconstruction theory is used to get the final reconstructed image on the low rank part with low and high resolution dictionary pair. Experimental results demonstrate that the proposed algorithm is robust to noise, gains clear visual appearance and high efficiency, and achieves a desirable improvement of 4 dB averagely compared with SPBSR