不同图像复杂度下超分辨率去噪算法研究  

Research on Super-Resolution Denoising Algorithms under Different Image Complexities

作　　者：钟小莉[1] 谢旻旻[1] ZHONG Xiao-li;XIE Min-min(Computer School of Qinghai Minzu University,Qinghai Xining 810007,China)

机构地区：[1]青海民族大学计算机学院,青海西宁810007

出　　处：《计算机仿真》2024年第12期312-316,共5页Computer Simulation

基　　金：基于立体图像智能分割技术应用的算法研究(2018XJY01)。

摘　　要：在处理高复杂度图像时,受各种噪声的影响,导致图像部分信息丢失,从而使得图像质量变差。为了有效提升图像质量,提出一种不同图像复杂度下超分辨率去噪算法。对于含有噪声图像的各个像素点邻域,采用多次马尔科蒙特卡洛(Markov Chain Monte Carlo,MCMC)随机采样搜索多个相似匹配块组,通过不同匹配组估计的一致性,获取调整相似块尺寸的判定依据,经过比较确定最佳相似块尺寸和对应的最优相似块组,引入修正的双向非局部算法获取无噪相似块估计,利用叠加的方法获取去噪后的图像。将高分辨率(High-Resolution,HR)图像退化为低分辨率(Low-Resolution,LR)图像,通过训练获取完备字典,将不同复杂度下的超分辨率重建问题展开稀疏表示,并对其求解;采用重叠技术缓解块效应,使用反向投影技术保证全局一致性,获取重建后的图像。实验结果表明,所提算法的图像峰值信噪比高,且结构相似度接近1,表明所提算法可以获取满意的去噪效果。When processing highly complex images,various noises can cause partial information loss,resulting in a decrease in image quality.In order to effectively improve image quality,a super-resolution denoising algorithm is proposed for different image complexities.Firstly,for the pixel neighborhood of noisy image,we used Markov Chain Monte Carlo(MCMC)random sampling to search multiple similar matching blocks.By estimating the consistency of different matching groups,we obtained the basis for adjusting the size of similar blocks.After comparison,we determined the best similar block size and corresponding optimal similar block group.Then,we introduced a modified bidirectional non-local algorithm to estimate noise-free similar block.At the same time,we used the superposition method to obtain the denoised image,so that the High-Resolution(HR)image can be degraded to Low-Resolution(LR)image.After that,we obtained a complete dictionary by training the image.Furthermore,the super-resolution reconstruction problem under different complexities was expanded into sparse representation and solved.Finally,we used the overlapping technology to alleviate blocking effect.Meanwhile,we used the reverse projection technology to ensure the global consistency,thus obtaining the reconstructed image.Experimental results prove that the proposed algorithm has high peak signal-to-noise ratio,and its structural similarity is close to 1,indicating that the algorithm can achieve satisfactory denoising effect.

关 键 词：不同图像复杂度下 超分辨率 去噪 稀疏表示 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]