导　　师： 练秋生

学科专业： 081001

授予学位： 硕士

作　　者： ;

机构地区： 燕山大学

摘　　要： 图像超分辨率重构技术是利用同一场景下一帧或一个序列低分辨率图像来估计一帧或一个序列高分辨率图像的数字信号处理技术。图像超分辨率重构技术能够突破图像采集系统的硬件限制,利用图像自身的冗余信息以及先验知识,通过软件方法提高图像质量。目前,超分辨率重构技术已经广泛应用于公共安全、遥测遥感、医学影像以及高清晰度电视等领域。 本文利用图像的稀疏性先验知识,针对单帧图像超分辨率重构问题进行研究,主要有以下三个方面: 首先,针对传统插值方式产生边缘模糊的缺点,提出了基于三次样条插值和方向插值融合的重构算法。利用图像在小波域的稀疏性区分图像的平滑区域与边缘,并利用图像梯度信息对边缘方向进行判断,对平滑区域和边缘区域分别采用三次样条插值和方向插值重构。实验结果表明融合插值的方法能够获得清晰的边缘。 其次,针对现有稀疏表示算法字典单一,不能最稀疏表示不同类型图像块的缺点,提出基于图像块分类稀疏表示的重构算法。将图像块分为平滑、不同方向的边缘和不规则结构多个类型,并训练各自对应的高、低分辨率字典进行重构。实验结果表明算法重构图像的视觉效果较好,并且算法速度有一定提高。 最后,将图像块分类稀疏表示理论应用到压缩传感图像重构当中,提出基于图像块分类字典和双树复数小波双层稀疏表示的压缩传感图像重构算法。先利用分类字典进行分块重构,然后利用图像在双树复数小波下的稀疏性,通过迭代收缩算法提高重构图像质量。实验结果表明了算法的有效性。 Image super-resolution reconstruction is a digital signal processing technique to estimate a high-resolution image /(or image sequence/) from a single image or an image sequence of the same scene. Image super-resolution reconstruction technique can break through the hardware limitation of image collection system, use the redundant information and prior knowledge of image itself to enhance image quality by software methods. At present, this technique has been widely used in public safety, telemetry remote sensing, medical imaging and high-definition TV, etc. This paper performs the researches on single frame image super-resolution reconstruction problems using image sparse prior knowledge, mainly in the following three aspects: Firstly, we propose the reconstruction algorithm based on fusion interpolation aiming to overcome the shortcomings of traditional interpolation methods which often cause blur effect in edges.The proposed algorithm firstly divides the image into smooth area and edges using its sparsity in wavelet domain and judes the edges’orientation using gradient priors, then uses spline interpolation for smotch aeras and directional interpolation for image edges.The experiment results show that the proposed algorithm can produce sharp edges. Secondly,because of the single dictionary can not sparsely represent the image patches of different types, we propose the reconstruction algorithm based on sparse representation of classified image patches. The proposed algorithm divides the image patches into smooth patches, edge patches of different orientation and irregular structure patches, and then trains their corresponding dictionary pairs of high and low resolution for image reconstruction. The experiment results show that the proposed algorithm can produce better visiual quality images and has faster speed. Finally, we apply the sparse representation theory of classified image patches to compress sensing image reconstruction, and propose the reconstruction algorithm based on double spa

关 键 词： 超分辨率 稀疏表示 融合插值 块分类 正交匹配追踪 双树复数小波

领　　域： [自动化与计算机技术] [自动化与计算机技术]