遥感影像超分辨率重建的字典学习类算法

近年来基于字典学习的超分辨率重建技术已成为图像处理领域的研究热点,相比基于重建的超分辨率方法,基于学习的方法充分利用了先验知识,在放大倍数较高时,仍可取得较好的效果,因此被公认为一种非常有前途的方法。本文对国内外已有的基于字典学习的超分辨率重建方法进行了系统研究,梳理了3种基于字典学习超分重建算法的基本原理及优缺点。此外,本文根据遥感影像的特点,使用同一数据源进行字典学习,利用不同字典学习算法分别生成高、低联合字典对,采用不同尺寸大小及缩放倍数的测试图像,进行超分辨率重建,对各种算法的重建性能、鲁棒性和复杂度进行综合分析,进一步研究了各种算法对遥感影像不同应用需求的适用性。

Super-resolution Methods Based on Dictionary Learning for Remote Sensing Images

In recent years, super-resolution reconstruction technology based on dictionary learning has obtained much attention and has been intensively studied.Compared with the super-resolution method based on reconstruction, the learning-based method makes full use of prior knowledge.This learning-based method can get better results when magnification is high, which has been recognized as an extremely promising method.The properties of the existed learning-based super-resolution reconstruction algorithms are firstly analyzed systematically.Then this paper reviews the theory of three learning-based algorithms and combs their advantages and disadvantages.Finally, according to characteristics of remote sensing image, the same data sources are used for dictionary learning.We select these three algorithms mentioned to generate high and low resolution joint dictionary and adopt test images of different sizes and zoom and complete reconstruction.The reconstruction performance, robustness and complexity of various algorithms are analyzed comprehensively by experimental results.What’s more, aiming at different application requirements of remote sensing image, the applicability of different algorithms is further studied.