摘要
针对静止场景中由于相机的不均匀抖动而产生的空间变化模糊图像,提出一种带有模糊核选择器的基于分块的全局图像复原算法.该算法将图像规则分块并用不同参数对各区块做模糊核辨识,得到各分割区块的一组模糊核估计;通过设计的残差评估的策略,从辨识到的模糊核集合中筛选出与各区块匹配的最佳模糊核估计;再通过迭代最小化目标函数求得最终的全局复原结果.实验结果表明,当图像中包含不同抖动模糊水平的子区域而且子区域间的轮廓划分较鲜明时,文中算法的复原效果尤佳.
Aiming at space-variantly blurred images caused by non uniform camera shake in the static scene, a partition-based global image-restoration algorithm with a blur kernel selector was proposed. The image was partitioned into several regular blocks, and different parameters were employed to identify the blur kernels of each block so that a set of blur-kernel estimations were obtained for each segmented block; through a designed strategy of residual error evaluation, the optimal kernel estimation toward each subregion was screened out from the identified blur-kernel set; afterwards, the final and global restoration result was acquired by iteratively minimizing an objective function. Experiments show that in the case of images that contain subregions, which are of different shake-blur levels and whose contour divisions between each other are relatively distinct, restoration effects of the algorithm are particularly good.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2012年第6期766-774,共9页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60972126)
国家自然科学基金重点项目(61032007)
国家自然科学基金联合基金(U0935002/L05)
北京市自然科学基金(4102060)
关键词
相机抖动分类法
空间变化
抖动模糊
分块法
图像复原
残差
taxonomy of camera shake
space-variant
shake blur
partition method
image restoration
residual error
