论文部分内容阅读
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。
In recent years, the technology of dictionary-based super-resolution reconstruction has become a research hotspot in the field of image processing. Compared with the reconstruction-based super-resolution method, the learning-based method makes full use of prior knowledge, and at high magnification, Get better results, it is recognized as a very promising method. This paper systematically studies the existing super-resolution reconstruction methods based on dictionary learning at home and abroad, and reviews the basic principles and advantages and disadvantages of three kinds of super-reconstruction algorithms based on dictionary learning. In addition, according to the characteristics of remote sensing images, the same data source is used for dictionary learning. Different dictionary learning algorithms are used to generate high and low joint dictionary pairs respectively. Test images with different sizes and magnifications are used for super-resolution reconstruction. The reconstruction performance, robustness and complexity of the algorithm are analyzed comprehensively, and the applicability of various algorithms to different application requirements of remote sensing images is further studied.