基于改进模糊参数估计的R-L运动图像复原算法研究

Research on R-L Motion Image Restoration Algorithm Based on Improved Motion Parameter Estimation

本文通过分析图像复原过程,提出了改进的在倒谱域中估计模糊参数的算法。首先,对运动模糊图像进行灰度变换计算,得到该图像的倒频谱后进行压缩运算和居中化的操作;其次,对得到的图像进行边缘检测运算、二值化处理;再次,再进行Radon变换,用来消除十字亮线引起的干扰信息,以正确估计模糊图像的模糊角度;然后,通过利用一阶差分自相关和倒谱三维图的峰值对称性,计算峰值并得到模糊尺度;最后,基于以上算法估计的模糊参数构造点扩散函数,使用Richardson-Lucy(R-L)算法对其进行多次迭代使运动模糊图像复原。实验表明,相比于其它算法,本文提出的算法不仅能够支持更大的模糊尺度估计范围,且在范围内运动模糊尺度的绝对误差在4个像素以内,运动模糊角度的绝对误差在0.17°以下。使用本文提出的改进的模糊参数估计的R-L算法,可以更快地迭代到最佳复原图像,获得细节清晰的图像。

This paper proposes an improved algorithm for estimating blur parameters in the cepstrum domain by analyzing the image recovery process.First,we need to perform grayscale transformation calculation on the motion blurred image,and then perform compression and centering operations after obtaining the cepstrum of the image.Secondly,we perform edge detection operation and binarization on the obtained image.Then,we perform Radon transform to eliminate the interference information caused by the bright cross line,so that the blur angle of the blurred image can be estimated correctly.In addition,we also compute the peaks and obtain the blur scale by exploiting the first-order difference autocorrelation and the peak symmetry of the cepstral 3D plot.Finally,we construct a point spread function based on the blur parameters estimated by the above algorithm,and use the Richardson-Lucy algorithm for multiple iterations to restore motion blurred images.Experiments show that the algorithm proposed in this paper not only supports a larger range of fuzzy scale estimation than other algorithms,but also the absolute error of motion blur scale within the range is within 4 pixels,and the absolute error of motion blur angle is below 0.17°.Using the improved R-L algorithm for blur parameter estimation proposed in this paper,it is possible to iterate faster to the best recovered image and obtain images with clear details.