Review of non-blind deconvolution image restoration based on spatially-varying PSF
-
摘要: 传统的图像复原一般认为点扩散函数(PSF)是空间不变的,实际光学系统由于受到像差等因素的影响,并非严格的线性空间不变系统,基于空间变化PSF的非盲去卷积图像复原法逐渐体现其优越性。空间变化PSF的非盲去卷积图像复原法先准确估计图像空间变化的PSF,再利用非盲去卷积算法对图像进行复原,有利于恢复出高质量图像。本文从算法的角度综述了近几年提出的基于空间变化PSF的非盲去卷积图像复原方法,并对比了基于强边缘预测估计PSF的非盲去卷积法、基于模糊噪声图像对PSF估计非盲去卷积法等算法的优缺点,各算法分别在PSF估计精确度、振铃效应抑制效果、适用范围等方面体现出各自的优劣。空间变化PSF的非盲去卷积图像复原法的研究,有利于推进图像复原技术向更高水平发展,使光学系统往轻小型化方向发展,从而在多个科学领域发挥其重要作用。Abstract: Traditional image restoration is generally considered that point spread function(PSF) is space-invariant.However, the actual optical system suffering from various optical aberrations can not be strictly linear space invariant.Non-blind deconvolution(NBD) algorithm of image restoration based on spatially-varying PSF(SVPSF) gradually embodies its superiority.NBD image restoration with SVPSF accurately estimates the spatially-varying PSF of the image at first, and then restores the image through NBD algorithm, which is conducive to the recovery of high quality images.From the perspective of algorithm, we review non-blind image restoration method proposed in recent years based on spatially-varying PSF, as well as compare merits and drawbacks among NBD algorithm based on PSF estimation using sharp edge prediction, NBD algorithm based on blurred/noisy image pairs, and so on.These algorithms reflect pros and cons respectively in PSF estimation accuracy, inhibitory effect of ringing artifacts, and the scope of application.The study of the NBD image restoration method based on SVPSF is beneficial to the development of image restoration technology to a higher level, which facilitates the optical systems to be smaller, so that it can play an important role in many scientific fields.
-
Key words:
- image restoration /
- spatially-varying PSF /
- non-blind deconvolution /
- PSF estimated
-
图 3 改进RL算法和能量最小化算法恢复图像
Figure 3. Recovered images by variant RL and energy minimization
图 5 Christian J.Schuler等人实验的恢复图像
Figure 5. Recovered image in the experiment by Christian J.Schuler,et al
表 1 5种算法实例对比结果
Table 1. Comparison results of five examples using different algorithms
PSF估计 去卷积 适用范围 优点 缺点 Joshi 强边缘预测法 Richardson-Lucy 运动、离焦以及相 机的固有属性 所引起的模糊 ①速度快 ②PSF估计较准确 ③亚像素分辨率下 复原图像 不适用于模糊核 为多峰的情况 Michal S orel 模糊/噪声图 像对估计法 改进的RL算法 运动模糊 ①比单幅图像复 原速度更快 ②振铃效应抑 制效果好 拍摄两幅不同曝 光度的图像不 易控制 Hao Lin 刀刃法+ 分段插值 最小二乘共轭梯度 像差引起的模糊 避免振铃效应 像质提高不明显 Christian J.Schuler 频域直接计算 Hyper-Laplacian先 验+demosaicing 运动模糊、 像差模糊 ①PSF估计快速有效 ②恢复图像全彩色 ③能用于单透镜 成像复原 单透镜复原效 果差一些，PSF 有待改进 Heide 一阶原始对 偶算法 一阶原始对 偶算法 像差引起的模糊 ①速度快，鲁棒性好 ②振铃效应抑制 效果好 ③可用于多种相 机图像复原 简单透镜系统的 图像复原效果 差一些 
下载: 导出CSV
-
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [41] [42] [43] -
下载:
图(6) / 表(1)
计量
- 文章访问数: 3967
- HTML全文浏览量: 1095
- PDF下载量: 1868
- 被引次数: 0
