一种基于BM3D算法改进的图像去噪方法  

An Improved Image Denosing Method Based on BM3D Algorithm

作　　者：昝明远 周先春 王心晔[1] 张浩瑀 ZAN Mingyuan;ZHOU Xianchun;WANG Xinye;ZHANG Haoyu(Changwang School of Honors,Nanjing University of Information Science&Technology,Nanjing 210044;School of Artificial Intelligence,Nanjing University of Information Science&Technology,Nanjing 210044;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing University of Information Science&Technology,Nanjing 210044)

机构地区：[1]南京信息工程大学长望学院,南京210044 [2]南京信息工程大学人工智能学院,南京210044 [3]南京信息工程大学江苏省大气环境与装备技术协同创新中心,南京210044

出　　处：《计算机与数字工程》2024年第2期528-534,共7页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:11202106,61302188);江苏省“信息与通信工程”优势学科建设项目资助。

摘　　要：在图像处理领域中,三维块匹配算法(BM3D)算法结合了空间域与频率域的相关特性,是当前图像去噪效果最好的算法之一。由于该算法存在计算耗时长、相似块提取不够准确、分割块内分布不同的相似特征无法提取等问题,论文提出基于Radon变换进行图像信息降维,并通过求解降维函数的方差值、以及方差最大值坐标定位进行相似块信息的判断,再计算各相似块相对参考块的像素均值比例,并放缩再用于去噪。实验结果显示,优化算法去噪图像相较于原算法去噪图像运行时间减短,图像质量提升。In the field of image processing,the Block-matching and 3D filtering(BM3D)combines relevant characteristics of spatial domain and frequency domain,and it is one of the best image denoising algorithms at present.Because the algorithm has time-consuming computation,inaccurate extraction of similar blocks and inability to extract similar features with different distributions in segmentation blocks,the paper proposes a method of dimensionality reduction of image information based on Radon transform,and judges similar block information by solving variance value of dimensionality reduction function and coordinating location of maximum variance.And then the pixel mean ratio of each similar block relative to the reference block is calculated and scaled for denoising.The experimental results show that compared with the original algorithm,the running time of the denoised image is shorter and the image quality is improved.

关 键 词：图像去噪 近似块匹配 BM3D RADON变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]