稀疏平滑特性的多正则化约束图像盲复原方法

Sparsity and Smoothing Multi-Regularization Constraints for Blind Image Deblurring

为了实现对线性空间不变的模糊图像的盲复原,提出了一种基于稀疏性和平滑特性的多正则化约束的模糊图像盲复原方法.首先,根据自然图像边缘的稀疏特性,运用了一种权重的全变差范数(weighted total variation norm,简称WTV-norm)对复原图像进行正则化约束;然后,从运动模糊的点扩散函数(motion point spread function,简称MPSF)的特性出发,提出一种能够适用于多种模糊情况的多正则化约束;最后,提出了一种改进的变量分裂(modified variable splitting,简称MVS)方法来得到清晰的复原图像,同时准确地估计出相应的模糊退化函数.大量的实验结果表明,该方法能够较好地复原多种不同类型的模糊(例如运动模糊、高斯模糊、均匀模糊、圆盘模糊).与近几年提出来的一些具有代表性的模糊图像盲复原方法相比,该方法不仅主观的视觉效果得到了较为明显的改进,而且客观的信噪比增量也增加了1.20dB^4.22dB.

In order to recover the linear spatial invariant blurred image blindly, a multi-regularization constraint-method for blind image resoration, which is based on the sparsity and the smoothing of the motion blur point spread function （MPSF）, is proposed. First, according to the sparsity of the edges in the natural image, a weighted total variation norm （WTV-norm） is applied to the restoration image. Next, inspired from the characteristics of the MPSF, multi-regularization constraints that handle a wide variety of blurring degradations, it is applied to the blur kernel. Finally, a modified variable splitting （MVS） method is proposed to restore the image and simultaneously estimate exactly the blur function. A large number of experimental results indicate that the proposed method can also undo a wide variety of blurring degradations （e.g. motion blur, Gaussian blur, uniform blur, disk blur）. In comparison with several recent representative image blind restoration methods, the subjective vision is not only more effective, but the ISNR also improved between 1.20dB and 4.22dB.