噪声模糊图像盲目反降晰的模糊核准确估计  

Accurate Kernel Estimation for Blind Deblurring of Noisy and Blurred Images

作　　者：孙士洁[1,2,3,4] 赵怀慈[1,3,4] 郝明国[1,3,4] 李波[1,2,3,4,5] 

机构地区：[1]中国科学院沈阳自动化研究所光电信息技术研究室,沈阳110016 [2]中国科学院大学,北京100049 [3]中国科学院光电信息处理重点实验室,沈阳110016 [4]辽宁省图像理解与视觉计算重点实验室,沈阳110016 [5]沈阳工程学院信息学院,沈阳110136

出　　处：《计算机辅助设计与图形学学报》2016年第5期813-820,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：总装国防预研项目

摘　　要：针对大多数先进的单幅图像盲目反降晰技术在噪声无法忽略时,仍不能很好地处理模糊核估计质量退化严重的问题,提出一种利用图像显著结构从单幅噪声模糊图像中准确估计模糊核的方法.首先通过降噪预处理对图像噪声进行抑制,利用基于全总变分模型的方法提取模糊图像的显著结构,进而运用梯度选择方法移除不利于模糊核估计的显著边缘,提高模糊核估计的鲁棒性;然后采取两阶段模糊核估计策略,运用基于图像显著结构模糊核估计方法和迭代支持域检测技术实现模糊核的准确估计;最后通过稀疏先验约束的非盲目图像解卷积方法完成最终的图像恢复.实验结果表明,与已有方法相比,该方法在合成和真实图像上都能更准确地估计出噪声模糊图像的模糊核,获得更好的复原图像质量,可有效地处理图像反降晰对图像噪声敏感问题,实现了噪声模糊图像模糊核的准确估计.Most state-of-the-art single image blind deblurring techniques can’t still handle perfectly theproblem that the quality of blur kernel estimate can be degraded dramatically when the input image noisecan’t be ignored. In this work, we present a new method for estimating an accurate blur kernel from a blurryand noisy image using salient image structure. First, we use denoising as a preprocess to remove the inputimage noise, and then compute salient structure of the denoised result based on the total variation （TV）model. We also apply a gradient selection method to remove those salient edges that have a possible adverseeffect on blur kernel estimation, thus improving the robustness of blur kernel estimation. Next, we adopt atwo-phase blur kernel estimation strategy to achieve an accurate kernel estimation by taking advantage of theblur kernel estimation method from salient structure and iterative support detection （ISD） technique. Finally,we choose to use the non-blind deconvolution method with sparse prior knowledge to attain the final latentimage restoration. Experiment results on synthetic and real world data show that our method produces moreaccurate blur kernels and higher quality latent images than previous approaches on noisy and blurry images.It handles effectively the truth that image deblurring techniques are very sensitive to noise, and estimates anaccurate blur kernel from a noisy and blurry image.

关 键 词：盲目图像反降晰 噪声模糊图像 显著结构 图像边缘 模糊核估计 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]