Accuracy Comparison between Learning Method and Signal Processing Method Using Iteration for Severely Blur Images
Accuracy Comparison between Learning Method and Signal Processing Method Using Iteration for Severely Blur Images
复制标题
DOI：
10.1145/3441369.3441381
复制
发表时间：
2020-11
期刊：
影响因子：
--
通讯作者：
Masahiro Goto;T. Goto
中科院分区：
文献类型：
--
作者：
Masahiro Goto;T. Goto
模糊是最常见的图像退化类型之一。当模糊函数（PSF）未知并且要恢复退化图像时，传统方法需要从单个输入图像估计两个未知量，即PSF及其理想图像。因此，交替执行PSF估计和理想图像估计处理的方法是成功的。另一方面，近年来，使用AI的盲图像恢复取得了显着进展，可以进行更清晰的估计。在本文中，我们比较了传统的迭代方法与人工智能方法，旨在提高图像包含大模糊，这是不期望在传统的测试图像的性能。
Blurring is one of the most common types of image degradation. When the blurring function (PSF) is unknown and a degraded image is to be recovered, the conventional method requires the estimation of two unknowns, which are the PSF and its ideal image, from a single input image. Thus, the method of performing alternating a PSF estimation and an ideal image estimation processing has been successful. On the other hand, blind image restoration using AI has made remarkable progress in recent years, enabling clearer estimation. In this paper, we compare the conventional iterative method with the AI method, and aim to improve the performance of images containing large blurring, which was not expected in conventional test images.