基于分数阶全变分正则化的超分辨率图像重建  

Super-resolution Image Reconstruction Based on Fractional Order Total Variation Regularization

作　　者：刘亚男[1] 杨晓梅[1] 陈超楠[1] 

机构地区：[1]四川大学电气信息学院,成都610065

出　　处：《计算机科学》2016年第5期274-278,307,共6页Computer Science

摘　　要：从退化的低分辨率图像重建得到高分辨率图像的本质是一病态逆问题,针对该问题,通过添加正则项进行处理。在使用传统的全变分(TV)的基础上,添加了分数阶全变分(FOTV)作为另一正则项来约束解空间。分数阶全变分正则项的使用可以更好地重建图像的细节纹理信息,弥补了全变分算子在平滑区域易出现阶梯效应的缺陷。利用交替方向乘子(ADMM)算法将问题划分为子问题,将全变分和分数阶全变分算子作为循环矩阵,通过傅里叶变换将其对角化,降低了计算的复杂程度。实验结果表明,与已有的方法相比,所提方法有效地避免了阶梯效应的产生,较好地保持了细节信息,并且具有更好的峰值信噪比(PSNR)和结构相似度(SSIM)。It is an ill-posed that a high resolution image is reconstructed from a degenerate low resolution image,and regularization is added to deal with the problem usually.In this paper,we introduced fractional order total variation（FOTV）as another regularization to constrain the solution space on the basis of traditional total variation（TV）operator.Detailed texture information of the image was better reconstructed by using FOTV regularization,and staircase effect was eliminated.Moreover,we divid the problem into sub-problems by alternating direction multiplier method（ADMM）,and total variation and fractional total variation operators were constructed as cyclic matrices.Then,these were diagonalized by Fourier transformation.Therefore,computational complexity is reduced.Experimental results show that compared to existing methods,the proposed model does not suffer from staircase.Furthermore,the proposed model can keep the details of the information and has better value of peak signal to noise ratio（PSNR）and similarity index measure（SSIM）.

关 键 词：超分辨率图像重建 全变分 分数阶全变分 交替方向乘子法 阶梯效应 纹理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]