贝叶斯框架下的模糊图像盲去卷积算法

为了去除图像模糊的同时,保持图像边缘等细节信息,需要对原始图像和点扩散函数进行准确的估计.在贝叶斯框架下,基于总变分模型,建立原始图像和点扩散函数的先验模型,同步估计原始图像和点扩散函数.对于总变分模型不可微分的问题,在不影响速度的前提下,用迭代重加权范数算法处理该问题.基于共轭分布理论,提出以伽马分布作为未知参数的先验模型,准确估计参数.实验表明该算法在对原始图像、点扩散函数和参数准确估计的基础上,成功地解决了模糊图像的盲去卷积问题,算法的速度和效果都得到了改进.与同类算法相比,本文提出的算法具有一定优势.

Bayesian Framework Based Blind Deconvolution Algorithm for Blurred Images

For the purpose of deblurring the blurred images without the loss of the detailed information as edges, it was necessary to estimate the original image and point spread functions accurately. This paper proposed a Bayesian framework based algorithm which used the total variation model to describe the original image. The total variation could preserve the edges of deblurred images, while it was non-differentiable. Therefore, we utilized the iteratively reweighted norm method to solve this problem. Based on the concept of the prior distribution, the Gamma distribution was introduced as the prior models of the unknown model parameters. The experimental results show the competitive performance of the proposed algorithm. With the accurate estimation of the original image and unknown model parameters, the blind devconvolution can be implemented successfully. The speed of the proposed algorithm and the results of the deconvolution are all improved obviously. Compared with the similar algorithm, the proposed algorithm has some advantages.