摘要
运动模糊图像的盲复原一直以来都是一个极具挑战的问题。为了能够准确地估计出运动模糊核(Motion Blur Kernel:MBK),进而得到高质量的复原图像,提出了一种基于正则化技术的多约束运动模糊图像盲复原方法。首先,为了能够准确地提取出图像中的大尺度边缘,提出了一种基于梯度选择的稀疏图像平滑方法;然后,在MBK的估计阶段,根据运动模糊核的内在特性,提出了一种多约束的正则化模型,同时结合提取的大尺度图像边缘,实现了对MBK的准确估计;最后,采用了半二次性的变量分裂策略对在模糊核估计阶段所提出的多约束正则化模型进行最优化求解,能够在准确估计MBK的同时得到高质量的复原图像。分别在人造的模糊图像和真实的模糊图像上进行了大量的实验,实验结果表明:提出的方法较近几年的一些代表性的较为成功的运动模糊图像盲复原方法相比,在主观的视觉效果和客观评价指标两方面都具有明显的改进。
Blind restoration of a motion-blurred image is a long-standing and challenging inverse problem. In order to estimate motion blur kernel （MBK） accurately and obtain a high-quality restoration image, a regularization-based multi-constraint blind restoration method for motion-blurred images is proposed. First, in order to extract the large-scale edges from the image accurately, a sparse image smoothing method, based on gradient selection, is proposed. Then, in the MBK estimation step, based on the inherent properties of the MBK, a multi-constraint regularization model, which combines the extracted large-scale image edges, is proposed. Finally, the multi-constraint regularization model, which is proposed in the MBK estimation step, is addressed by using a half-quadratic variable splitting scheme. Extensive experiments are performed on both synthetic blurred images and real-life blurred images. Experimental results indicate that in comparison with several recent successful representative image blind restoration methods, the proposed method is an improvement not only in terms of subjective vision, but also in terms of objective numerical measurement.
出处
《红外技术》
CSCD
北大核心
2017年第12期1098-1106,共9页
Infrared Technology
基金
陕西省科技计划(工业攻关)项目(2014K05-22)
关键词
运动模糊图像
盲复原
大尺度图像边缘
运动模糊核
多约束正则化模型
motion blurred image, blind restoration, large scale image edges, motion blur kernel, multi-constraint regularization model
