基于多级残差反馈的有限脊波变换图像去噪算法  

Finite Ridgelet Transform Image Denoising Algorithm Based on Multi-level Residuals Feedback

作　　者：李思桐 刘光前 葛唯益[2] 纪建[1] 苗启广[1] LI Sitong;LIU Guangqian;GE Weiyi;JI Jian;MIAO Qiguang(School of Computer Science and Technology, Xidian University, Xi'an 710071, China;Science and Technology on Information Systems Engineering Laboratory, Nanjing 210007, China)

机构地区：[1]西安电子科技大学计算机科学与技术学院,西安710071 [2]信息系统工程重点实验室,南京210007

出　　处：《指挥信息系统与技术》2019年第5期47-53,共7页Command Information System and Technology

基　　金：国家自然科学基金(61772396)资助项目

摘　　要：为了充分利用残差图像中的结构信息,提出了一种基于多级残差的有限脊波图像去噪算法,以提升有限脊波变换的去噪性能.为了增强和丰富去噪图像的边缘和纹理信息,提高去噪图像的视觉效果,该算法先对有限脊波去噪后的残差图像反复进行块匹配三维滤波(BM3D)去噪和高斯滤波,从而获得图像的结构信息,再将其作为补偿图像反馈给去噪图像,并提出一种基于图像相关性的最优迭代停止准则.最后,通过多组仿真试验验证了该迭代停止准则和去噪算法的可行性和有效性.In order to use the structure information in the residual image,a finite ridgelet transform image denoising algorithm based on multi-level residuals feedback is proposed to improve the denoising performance of the finite ridgelet transform. To enhance and enrich the edge and texture information of the denoised image, and improve the visual effect of denoised image, the block matching and 3D filtering (BM3D) and Gaussian filtering are used repeatedly to get the structure information, and it is fed back to the denoised image as a compensation image. At the same time, a new optimal iterative stopping criterion based on image correlation is proposed. Finally, the feasibility and effectiveness of the iterative stopping criterion and the denoising algorithm are verified by a number of simulation experiments.

关 键 词：图像去噪 有限脊波变换 块匹配三维滤波 残差图像 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]