基于超拉普拉斯模型的运动去模糊算法

Motion deblurring based on hyper-Laplacian model

通过分析自然图像梯度统计的长尾分布特性,提出一种基于超拉普拉斯模型的图像运动去模糊算法。该算法使用分步交替迭代最小化办法优化能量方程,并在优化过程中采用了查询表法快速求解图像反卷积。实验结果表明,该算法可以在数秒时间内处理完成一幅中等大小的图像,并获得较优的复原结果,其复原效果和对振铃现象的抑制均优于维纳滤波法和Richardson-Lucy复原法。

This paper first analyzed the heavy-tailed distribution of gradients in natural scenes and proposed an algorithm about motion deblurring based on hyper-Laplacian moldel.It adopted an alternating minimization scheme to optimize the energy equation.Used a look-up table（LUT） for rapidly solving the image deconvolution.Our method can yield good result with medium size pictures in seconds.Compared with the Wiener and Richardson-Lucy methods,our method yields the best overall performance both in the restoration effect and suppression of ringing.