结合基于梯度的振铃评价算法的总变分最小化图像分块复原法

A Total-Variation Majorization-Minimization Sectioned Restoration Algorithm with Gradient Ringing Metric Image Quality Assessment

为了消除退化函数随空间变化发生变化模糊图像分块复原法子块之间的不平滑拼接缝,提出了一种结合了基于梯度的振铃评价算法梯度振铃评价(GRM)的总变分(TV)最小化分块复原法。根据图像分布及退化类型将模糊图像划分为矩形、环形或其他形状的子块,图像子块之间要留有一定的重叠区;然后对每一个图像子块进行复原,GRM方法是基于图像梯度结构相似度的图像质量评价算法,以GRM作为TV复原算法迭代过程中的收敛条件,可以更好地控制复原图像的振铃;最后去除复原图像子块含振铃波纹的重叠区,拼接得到完整图像。并以矩形分块及环形分块为例,证明该方法可以很好地抑制图像边界振铃效应,克服分块复原法本身的缺陷,得到拼接平滑的完整图像。

For eliminating the ringing artifacts between the sub-frames of the sectioned restoration algorithm for images with space-variant point spread function （SVPSF）, the paper introduces a sectioned restoration algorithm, which bases on total variation （TV） majorization-minimization restoration algorithm and gradient ringing metric （GRM） image quality assessment approach. Firstly, the SVPSF-blurred image is divided into rectangular sections, circular sections or any other, which relies on the distribution of the degradation function, with some overlapped- regions. Then, each sub-section is restored by TV restoration algorithm with GRM as the convergence limit of restoration iteration. The GRM method is helpful to identify ringing of restored image, which relies on the similarity of the gradients of two images. After removing the overlapped regions, the sub-frames are spliced together to construct the composite full image. Taking the restorations of the rectangular-section and circular-section SVPSF- blurred images as examples, the paper proves that the algorithm is good at suppressing ringing artifacts. Consequently a better image with smooth splicing is obtained. The drawback of the sectioned restoration algorithm is overcomed.