利用低秩先验的噪声模糊图像盲去卷积

发布时间：2019-02-22 12:08

【摘要】：单幅图像盲去卷积的目的是从一幅观测的模糊图像估计出模糊核和清晰图像。该问题是严重病态的,尤其是观测图像中噪声不可忽略时更具挑战性。该文主要针对如何有效利用低秩先验约束进行噪声模糊图像盲去卷积问题,提出一种在交替最大后验(MAP)估计框架下利用低秩先验约束的单幅噪声模糊图像盲去卷积方法。首先,在估计中间复原图像时,利用低秩先验约束对复原图像中的噪声进行抑制。然后,采用降噪后的中间复原图像估计模糊核,得到更好质量的模糊核估计。迭代上述两个操作获得最终可靠的模糊核估计。最后,根据所估计的模糊核,通过非盲去卷积方法复原出清晰图像。实验结果表明:所提方法在定量和定性评价指标上优于已有的代表性方法。
[Abstract]:The purpose of blind deconvolution of a single image is to estimate the blurred core and clear image from an observed blurred image. The problem is seriously ill-posed, especially when the noise in the observation image can not be ignored. In order to solve the problem of blind deconvolution of noise blur images using low rank prior constraints effectively, this paper presents a blind deconvolution method for noise blur images with low rank prior constraints under the framework of alternating maximum a posteriori (MAP) estimation. Firstly, the noise in the reconstructed image is suppressed by using a low rank priori constraint in the estimation of the intermediate restoration image. Then, a better quality fuzzy kernel estimation is obtained by using the de-noised intermediate restoration image to estimate the blur kernel. The final reliable fuzzy kernel estimation is obtained by iterating the above two operations. Finally, according to the estimated fuzzy kernel, the clear image is reconstructed by non-blind deconvolution. The experimental results show that the proposed method is superior to the existing representative methods in quantitative and qualitative evaluation.
【作者单位】：中国科学院沈阳自动化研究所光电信息技术研究室;中国科学院大学;中国科学院光电信息处理重点实验室;辽宁省图像理解与视觉计算重点实验室;沈阳工程学院信息学院;
【基金】：辽宁省教育厅科研项目(L2015368)~~
【分类号】：TP391.41

【相似文献】