去运动模糊一直是计算机视觉领域中面向画质增强的一个热点研究方向。模糊核的估算是去运动模糊中的关键问题。提出一种新的思路,即首先将模糊图像按照模糊核的相似度进行图像分割,再对分割后的图像应用空间不变去模糊算法。本文方法主要包含以下几个步骤:分离输入图像中的光照、颜色和纹理信息;分割图像;分区域估算模糊核,计算重叠区域模糊核,并根据计算出的模糊核进行分区域单核去模糊;利用重叠区域整合拼接去模糊结果并还原光照和颜色信息。实验结果表明本文方法比基于单核的去运动模糊算法效果要好。

Motion de-blurring is a heated topic in the area of image quality enhancement, where kernel estimation is the key challenge. In this paper, we propose a new approach, which first performs segmentation on the input image and then deblur different regions. The method described in this paper includes the following steps: image decomposition, image segmentation, region-based kernel estimation and non-blind de-convolution,as well as image reconstruction by use of overlapping regions. Experimental results show that our method can outperform single kernel motion de-blurring methods in some cases.