激光与光电子学进展, 2018, 55 (4): 041015, 网络出版: 2018-09-11
基于l 1/l 2的高低阶全变差运动模糊图像盲复原方法
Blind Recovery Method of Motion Blurred Image Based on Combining l 1/l 2 Norm with High Order and Low Order Total Variation
图像处理 图像盲复原 去模糊 l1/l2范数 高低阶全变差 分裂Bregman迭代 image processing image blind recovery deblurring l1/l2 norm combining high order and low order total variation split Bregman iteration
摘要
为了实现运动模糊图像的盲复原,提出了一种基于l 1/l 2的高低阶全变差图像盲复原方法。利用具有更强稀疏表达能力的l 1/l 2范式正则化先验项,加入高低阶混合全变差正则化模型。高阶全变差正则化模型可以抑制图像非边缘部分可能出现的阶梯及振铃效应,低阶全变差正则化模型可以保护自然图像的边缘稀疏特性。分别给出了清晰图像和模糊核的求解算法,两者的求解过程采用分裂Bregman迭代算法将目标函数分裂成多个子问题进行优化求解。实验结果表明,提出的方法能够很好地抑制振铃效应并保护图像的边缘细节,通过与其他盲复原方法进行比较,在视觉质量与客观质量评价上均说明本文算法具有更好的稳健性。
Abstract
In order to realize the blind recovery of motion blurred image, we present a blind recovery method based on combining l1/l2 norm with the high order and low order total variation. We adopt the ratio of l1/l2 norm regularization prior item which has high sparse expression ability, and add the high order and low order total variation regularization item. High order total variation regularization model can suppress the ladder effect and ringing effect that may occur in the region of non-edges. Low order total variation regularization model can protect the sparse feature of natural image edges. The solution of high-quality image and the solution of blurred kernel are given respectively, both of which employing Bregman iterative algorithms to split the objective function into multiple sub-problems. The experimental results show that the proposed method can restrain ringing effect and protect the image edge details. The robustness of proposed algorithm is better in the visual quality and objective quality evaluation comparing with other methods for blind recovery of motion blurred images.
王灿, 杨帆, 李靖. 基于