数字图像空间分辨率改善的插值-模拟采样迭代方法

An Iterative Method of Interpolation and Simulated Sampling for Spatial Resolution Improvement of Digital Images

在遥感图像和计算机视觉的一些应用中，需要从已有的一些低分辨率图像来获得更高分辨率的图像．已有一些方法都是针对大小完全相同和图像之间不存在局部的几何畸变（或假设如此）的图像，而对从大小不同或分辨率不同的图像和图像之间存在相对的几何畸变等更普遍存在的实际图像来重构更高分辨率图像的问题未加涉及．本文用从带限信号的非均匀采样进行信号重构的迭代方法和凸集投影方法出发，导出了一种应用范围更加广泛的从多幅空间分辨率低的图像重构更高空间分辨率图像的一般性空间域插值一模拟采样选代方法，并提出了凸集投影的变权选代解法．该选代方法可以对付图像之间存在相对的几何畸变、辐射亮度差异、空间分辨率差异及图像存在噪声等情形实验表明，该迭代解法具有很好的收敛性和很好的收敛效果．

Spatial resolution improvement of digital images has significant applications in remote sensing and computer vision. Some former algorithms were (or supposed to be) almost for multiframe images of same spatial resolution and without geometric distortion. In this paper, a general method of interpolation and simulated sampling for spatial resolution improvement from multiframes of undersampled digital images is proposed based on the iterative metheds for signal reconstruction fain nonuniform sampling and the methods of projection onto convex sets (POCS). The method can tackle multiframe images of various spatial resolutions,different image radiance, ralative geometric distortion, additive random noise, and so on. An idea of varying weights in iteration of parallel POCS methods is also discussed. Some experiments with resolution test pattern and multiangular remote sensing images show the convergence and effectiveness of the algorithms.