基于振铃约束的全变差正则化图像去模糊算法

Image deblurring based on ringing constraint with total variation regularization

当前去模糊方法只利用图像单一的稀疏特性作为先验信息,忽略了伪边缘(如振铃瑕疵)对模糊核估计的影响,导致其去模糊性能不佳。本文充分利用复杂结构图像的先验信息,设计了振铃约束下的全变差正则化图像去模糊算法。首先,利用多分辨率图像金字塔策略建立多层图像模型,通过对比模糊图像和潜在清晰图像来获得振铃先验信息。其次,将振铃正则约束项融入全变差方法,构建多正则项去模糊模型,然后利用变量分离法将去模糊模型转化为多函数优化问题。最后,利用一阶原始对偶算法,根据低分辨率到高分辨率的顺序,对模糊核和原始图像完成计算,获取重构目标。实验结果表明:较当前图像去模糊技术而言,所提算法具备更为理性的去模糊效果,所复原的图像呈现出更高的峰值信噪比和结构相似度,可以更好地保持图像边缘与纹理信息。

The current defuzzy method which only uses the sparse feature of the image as the prior information has poor defuzzy performance induced by ignoring the effect of false edges(such as ring defects)on the point spread function estimation.A regularized image deblurring algorithm with total variation under the constraint of ringing is designed based on the prior information of the complex structure image.Firstly,the multi-resolution image pyramid strategy is adopted to build a multi-layer image model,and the prior information of ringing is obtained by comparing the blurred image with the potentially clear image.Secondly,the ringing regularization constraint term is integrated into the total variation method to build a multi-regularization deblurring model,and then the variable separation method is utilized to transform the deblurring model into a multi-function optimization problem.Finally,the firstorder original-dual algorithm is employed to solve the Point Spread Function(PSF)and clear image in the order from low resolution to high resolution.Experimental results show that compared with the current image deblurring technology,the proposed algorithm has a more rational deblurring effect,and the recovered image shows higher peak signal-to-noise ratio and structure similarity,which can better preserve the image edge and texture information.