改进的广义高斯分布与非局部均值图像去模糊

Improved image deblurring based on combination of generalized Gaussian distribution and nonlocal-means

为了改善常规算法不能保留图像边缘细节信息的缺陷,获得更好的图像去模糊效果,在非局部均值图像复原算法的基础上提出一种新的基于广义高斯分布与非局部均值的去模糊算法。先对模糊图像进行小波变换,然后应用极大似然估计的方法以及经典的Newton-Raphson算法来估计出广义高斯分布模型的尺度参数和形状参数,利用这两个参数改进原始的单一根据指数函数的衰减速度和局限于一个参数来求图像权值的方法。在多个典型图像上的测试结果表明,改进算法后的图像去模糊化效果比原始的NL-means方法更优越,具有很好的应用前景。

In order to improve the traditional algorithm can＇t reserve the detail information of image defects,and obtain better image to the blurring effect,this paper proposed a new image deblurring algorithm based on generalized Gaussian distribution and NL-means.Firstly,it proposed the blurred image with the wavelet transform,and then estimated scale parameter and shape parameter of generalized Gaussian distribution model with OMLE and classical Newton-Raphson algorithm.It used these two parameters were used to improve initial image weights calculating method which was singly judged by decay rate of exponential function and restricted with only one algorithm.Experimental results on several typical images indicate that the proposed algorithm gives superior effects to initial NL-means and has a good application prospects.