改进正则项的图像盲恢复方法

Image blind deconvolution approach with modified regularization term

图像恢复是一个反卷积过程,这一过程通常是病态的,其中的盲恢复是一个最常见也最具挑战性的问题。由于盲恢复过程中缺乏点扩散函数的相关先验信息,使得这个过程变得更为复杂。为了保证在得到光滑图像的同时也可以很好地保持图像的边缘信息,本文提出了一个改进的全变分正则项的盲恢复模型,并结合分裂Bregman算法对模型进行了求解。数值计算中采用了快速傅里叶变换和shrinkage公式来降低计算复杂度。数值实验分别对模糊图、含有噪声和高斯模糊的灰度图进行了处理,得到了满意的结果。

Image restoration is a deconvolution process, which is usually morbid. Blind deconvolution is one of the most common and challenging problems. Without priori knowledge on point spread function, the process is therefore more complex, To preserve the edge information of an image as well as its smoothness, we present a blind deconvolution model with a modified total variation regularization term. We also solve the model with split Bregman iteration algorithm. Fast Fourier transform and shrinkage formula are applied in numerical calculation to reduce its computational complexity. We apply the model to the processing for a blurry image and a greyscale image with noise and Gaussian blur in numerical experiment, and then obtain satisfactory results.