﻿ 倍增比自适应的图像超分辨率重建
 计算机系统应用  2018, Vol. 27 Issue (12): 150-155 PDF

KUANG Qi-Gang, LIU Hao, WU Le-Ming, ZHANG Xin-Sheng, SUN Xiao-Fan
School of Information Science and Technology, Donghua University, Shanghai 201620, China
Foundation item: Natural Science Foundation of Shanghai (18ZR1400300)
Abstract: In recent years, the image super-resolution reconstruction has always been a hot research field, but the corresponding research results about arbitrary-scale-ratio super resolution are still rare. Under high scale ratio, the image resolution will become lower, and it is difficult for human eyes to recognize such image content. With the advancement of technology, machine vision has been used to recognize the images with very low resolution, and the research on arbitrary-scale-ratio super resolution has become increasingly important. Through testing various representative super-resolution algorithms, this study proposes a scale-adaptive super-resolution reconstruction algorithm according to a full-scale quality sum criterion after performing extensive arbitrary-scale-ratio analysis on image super resolution. Experimental results show that the proposed algorithm can achieve better overall-reconstruction performance within the whole scale range.
Key words: super resolution     arbitrary scale ratio     scale adaptive     full-scale quality     reconstruction algorithm

1 序言

Jiang等人在锚定邻域回归(ANR)方法的基础上, 提出了一种称为局部正则化锚定邻域回归(LANR)的快速图像SR方法, 应用局部约束来选择相似的字典原子, 通过在学习简单回归函数之前引入这种灵活性, 所提出的方法能够快速生成具有锐利边缘和丰富纹理的自然外观结果[9]. Jiang等人在LANR方法的基础上提出了一种新的局部正则化锚定邻域回归非局部均值(LANR+NLM)的图像SR方法, 并根据其与输入LR的相关性为每个字典原子赋予不同的自由度[10]. 尽管图像超分辨率重建是近年来的研究热点, 但是仍然还存在一些挑战性的问题[11].

2 任意倍增比的分析

 图 1 任意倍增比重建的折线图

3 任意倍增下的算法机制分析

3.1 全局回归机制

 图 2 倍增比为12时 barbara的重建图像

 $\begin{array}{*{20}{c}}{min}\\\beta \end{array}\left\| {yF - {N_l}\beta } \right\|_2^2 + \lambda {\left\| \beta \right\|_2}$ (1)

 $\beta = {(N_l^{\rm{T}}{N_l} + {{\lambda I}})^{ - 1}}N_l^{\rm{T}}yF$ (2)

 ${{x}} = {N_h}\beta$ (3)

3.2 局部正则化锚定邻域回归

 $N_{i,j}^L = \left\{ {D_L^K} \right.{\} _{k \in {C_K}\left( {D_L^j} \right)}}$ (4)

 $\widehat {{w_i}} = \begin{array}{*{20}{c}}{\arg min}\\{{w_i}}\end{array}\left\| {{y_i} - N_{i,j}^L{w_i}} \right\|_2^2 + {\lambda _1}\left\| w \right\|_2^2$ (5)

 ${w_i} = {(N_{i,j}^{LT}N_{i,j}^L + {\lambda _1}U)^{ - 1}}N_{i,j}^{LT}{y_i}$ (6)

 ${U_{kk}} = {g_{i,k}},k = 1,2, \cdots ,k.$ (7)

4 倍增比自适应策略

 ${{Q}} = \mathop \sum \limits_{k = 2}^{\rm{X}} {\rm{PSNR}}\left( k \right)$ (8)

5 实验结果

6 结论

 [1] 李云飞, 符冉迪, 金炜, 等. 多通道卷积的图像超分辨率方法. 中国图象图形学报, 2017, 22(12): 1690-1700. [2] Lu HY, Wei JB, Wang LZ, et al. Reference information based remote sensing image reconstruction with generalized nonconvex low-rank approximation. Remote Sensing, 2016, 8(6): 499. DOI:10.3390/rs8060499 [3] Ding M, Fan GL, Zhang X, et al. Structure-guided manifold learning for video-based motion estimation. Proceedings of 2012 19th IEEE International Conference on Image Processing. Orlando, FL, USA. 2013. 1977–1980. [4] 刘刚. 图像及视频序列超分辨率技术研究[博士学位论文]. 长春: 中国科学院长春光学精密机械与物理研究所, 2012. [5] Yang JC, Wright J, Huang TS, et al. Image super-resolution via sparse representation. IEEE Transactions on Image Processing, 2010, 19(11): 2861-2873. DOI:10.1109/TIP.2010.2050625 [6] Zeyde R, Elad M, Protter M. On single image scale-up using sparse-representations. In: Boissonnat JD, Chenin P, Cohen A, et al, eds. Curves and Surfaces. Berlin, Heidelberg: Springer, 2012. 711–730. [7] Timofte R, De V, Van Gool L. Anchored neighborhood regression for fast example-based super-resolution. Proceedings of 2013 IEEE International Conference on Computer Vision. Sydney, NSW, Australia. 2013. 1920–1927. [8] Chang H, Yeung D, Xiong YM. Super-resolution through neighbor embedding. Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington, DC, USA. 2004. 275–282. [9] Jiang JJ, Fu JC, Lu T, et al. Locally regularized anchored neighborhood regression for fast super-resolution. Proceedings of 2015 IEEE International Conference on Multimedia and Expo. Turin, Italy. 2015. 1–6. [10] Jiang JJ, Ma X, Chen C, et al. Single image super-resolution via locally regularized anchored neighborhood regression and nonlocal means. IEEE Transactions on Multimedia, 2017, 19(1): 15-26. DOI:10.1109/TMM.2016.2599145 [11] Zhao Y, Wang RG, Jia W, et al. Local patch classification based framework for single image super-resolution. arXiv:1703.04088, 2017.