摘要
结构组稀疏表示(structural group sparse representation,SGSR)算法对结构组的估计值进行奇异值分解得到字典,然后用Split Bregman Iteration算法求解优化模型得到稀疏解,最后借助字典和稀疏解来修复图像。该算法在一定程度上解决了传统稀疏表示算法忽略图像块之间相似性导致重构图像的结构和纹理不够自然的问题。但该算法中,结构组的估计值采用双线性插值算法得到,因此对块状缺失图像的修复效果一般。为了更准确地计算结构组的估计值,提出用Criminisi算法代替双线性插值算法,并由此时的估计值生成更合理的字典和稀疏解,得到重构的结构组,进而更准确地修复图像。实验数据表明,与SGSR算法相比,所提出的算法在峰值信噪比和相似结构性指数上分别平均提高了2.66 dB和0.0017,且在结构和纹理上取得了更自然的主观视觉效果。
The structural group sparse representation(SGSR)algorithm carries out singular value decomposition(SVD)on the estimate of a structural group to obtain the dictionary,then utilizes split Bregman iteration(SBI)algorithm to solve the optimization model for sparse coefficients,and finally adopts the dictionary and the coefficients to repair an image.In some sense,this algorithm solves the problem that the traditional sparse representation algorithm ignores the similarity between image patches,which will result in the fact that structures and textures in a reconstructed image are not natural enough.As the bilinear interpolation(BI)algorithm is used to calculate the estimate of a structural group,the SGSR algorithm does not fix the missing patch well.In this paper,in order to get the estimate of a structural group more accurately,we exploit the Criminisi algorithm to take the place of BI.It can obtain a more reasonable dictionary and coefficients from the estimate,and then reconstruct the structural group.Therefore,a better repaired image is produced.Experiment shows that compared with SGSR algorithm,the proposed algorithm is improved by 2.66 dB and 0.0017 respectively in terms of peak signal-to-noise ratio(PSNR)and similar structural index(SSIM),and it achieves more natural visual effects on textures of the reconstructed image.
作者
王君
唐贵进
刘小花
崔子冠
WANG Jun;TANG Gui-jin;LIU Xiao-hua;CUI Zi-guan(Jiangsu Key Laboratory of Image Processing and Image Communication,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)
出处
《计算机技术与发展》
2020年第3期24-29,共6页
Computer Technology and Development
基金
国家自然科学基金(61501260):江苏省研究生科研与实践创新计划项目(KYCX17_0776)
南京邮电大学科研基金项目(NY219076)。
关键词
图像修复
稀疏表示
字典学习
结构组
image inpainting
sparse representation
dictionary learning
structural group
