基于双L_0稀疏先验的图像运动去模糊

IMAGE MOTION DEBLURRING BASED ON DOUBLE L_0 SPARSE PRIORI

目前图像的运动去模糊方法在处理较复杂的运动模糊时难以得到理想的效果,其原因之一是这些方法普遍只考虑图像梯度的稀疏性,忽略了模糊核的稀疏性。针对这一不足提出一种新的双L_0正则约束的运动模糊去除方法,该方法同时对自然图像梯度和模糊核使用L_0正则约束,结合半正定二次分裂最小化的方法进行求解优化,实现自然模糊图像梯度和模糊核均稀疏下的模糊核估计,并进一步使用L_(0.5)超拉普拉斯正则约束项恢复最终图像。实验发现,该方法可以较好地去除单幅图像较复杂的运动模糊,更好地克服了估计的模糊核中存在的噪点和错误,得到较现有方法更加理想去模糊效果。

Existing image motion deblurring methods cannot obtain ideal results when dealing with the complex motion blurs.One of the reasons is that they generally only consider the sparsity of image gradients but ignore the sparsity of blur kernel.To overcome this limitation, this paper presents a new motion deblurring method with double L0 regular constraints,which applies the L0 regular constraints to both the natural image gradients and blur kernel,by combining the semi-definite quadratic splitting minimisation method it carries out the solution optimisation and realises the blur kernel estimation under the conditions of natural blurred image gradients and average sparsity of blur kernel. It further adopts a hyper-Laplacian term with L0.5 regular constraint to restore the final deblurred image.Experiment finds that the proposed method can well remove the rather complex motion blur of single image and better overcome the estimated noise and errors in blur kernel,and consequently obtains a more ideal motion deblurring effect than existing methods.