压缩感知在超分辨率图像重构技术中的应用研究

发布时间：2018-10-08 07:23

【摘要】：随着多媒体技术的不断发展,人们对于图像质量的要求越来越高。超分辨率技术在固定传感器精度的情况下实现了图像质量的提升,逐渐成为研究热点。近年来,基于信号稀疏表示的压缩感知理论在图像去噪,雷达成像等图像处理领域均取得了显著进展,利用压缩感知理论解决图像超分辨率重构问题,突破了传统超分辨率方法基于先验信息约束的固有局限,具有极高的研究价值。本文以压缩感知及其在超分辨率重构中的应用为研究主线,对单帧图像超分辨率重建,多帧图像超分辨率重建中的关键问题展开研究。针对重构图像边缘模糊问题,本文提出了基于压缩感知的单帧图像超分辨率重构框架,采用训练的过完备字典代替传统小波基,提高了图像稀疏表示性能。同时引入了过完备字典与观测矩阵联合训练的迭代优化方法,进一步降低观测矩阵与稀疏基之间的相关性,较好地恢复了图像细节信息。在多帧超分辨率重建中,本文首先分析图像退化理论模型,阐述了低分辨率图像序列之间具有的高相似性及信息冗余,在此基础上,本文提出将分布式压缩感知理论中的联合稀疏模型应用到多帧图像超分辨率重构中,实验表明,采用本文方法在减少数据量的情况下保证了图像的重构质量。
[Abstract]:With the continuous development of multimedia technology, people demand more and more high image quality. Super-resolution technology has achieved the improvement of image quality under the condition of fixed sensor precision, and has gradually become a research hotspot. In recent years, the theory of compressed sensing based on sparse signal representation has made remarkable progress in image processing such as image denoising, radar imaging and so on. The problem of image super-resolution reconstruction is solved by using the theory of compressed sensing. It breaks through the inherent limitation of the traditional super-resolution method based on the prior information constraint and has high research value. In this paper, compression perception and its application in super-resolution reconstruction are the main research thread, and the key problems in single-frame image super-resolution reconstruction and multi-frame image super-resolution reconstruction are studied. In order to solve the edge blur problem of reconstructed image, this paper proposes a single frame image super-resolution reconstruction framework based on compression perception. The trained over-complete dictionary is used to replace the traditional wavelet basis to improve the performance of image sparse representation. At the same time, an iterative optimization method of joint training of overcomplete dictionaries and observation matrices is introduced to further reduce the correlation between observation matrices and sparse bases and to restore the image details. In the multi-frame super-resolution reconstruction, the theoretical model of image degradation is first analyzed, and the high similarity and information redundancy among low-resolution image sequences are expounded. In this paper, the joint sparse model of distributed compression perception theory is applied to the super-resolution reconstruction of multi-frame images. The experiments show that the proposed method can guarantee the quality of image reconstruction under the condition of reducing the amount of data.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】