【作者】 李娟娟；

【导师】 李小红；

【作者基本信息】 合肥工业大学 ， 通信与信息系统， 2013， 硕士

【摘要】 图像超分辨率重建是指将模糊、降质的低分辨图像按照一定的算法进行处理，以获取清晰、高质量的高分辨率图像，它是图像处理、计算机视觉和模式识别等领域的研究热点问题，在天文学、遥感成像、军事、医疗等领域占有重要的地位。受压缩感知理论启发而提出的基于稀疏表示的超分辨率重建算法，在重建的效果和算法的复杂度方面要优于传统的算法，是目前广大学者所研究的热点之一，该算法的研究对进一步推动超分辨率重建技术的理论研究和实际应用具有重要的意义。本文主要围绕图像超分辨重建和图像的稀疏表示在图像超分辨率重建中的应用两方面进行了深入的研究，并对现有的算法进行改进创新。首先，本文对常见的图像超分辨率重建算法和稀疏表示理论进行了研究，针对图像超分辨率重建算法，文中将其分成了基于重建和基于学习两大类，对各算法进行了详细的介绍并分析了优缺点。针对稀疏表示理论，本文着重研究了欠定问题的解法和过完备冗余字典的设计问题，并将其应用到了图像超分辨率重建中。其次，本文在传统的基于稀疏表示的超分辨率重建的基础上，对过完备字典的构造和非局部相似性理论进行了研究，对重建过程中所需的字典构造方法和图像重建方法进行了优化改进，提出了一种新的基于聚类的单帧图像超分辨率重建算法。该方法首先从高分辨率样本图像中学习一个结构聚类型的高分辨率字典；然后利用迭代收缩算法优化目标方程来求得高分辨率图像的表示系数；最后通过学习到的高分辨率字典对低分辨率图像进行重构。最后，本文对该算法在无噪和有噪的环境下分别进行了多组仿真实验，并给出了仿真实验结果，结果表明该方法可以有效地提高单帧图像超分辨率重建的效果，并且降低了图像重建时的算法复杂度。更多还原

【Abstract】 Image super-resolution is referring to process low-resolution images by somealgorithms, in order to get high-resolution images. It is one of the most populartopics in many fields, such as images processing, computer vision andpatternfrecognition, etc. Image super-resolution reconstruction algorithms based onsparse representation which were inspired by compressed sensing have achievedbetter reconstruction results and complexity of the algorithm than classic methods.The research of the algorithms has significance in the theoretical research and thepractical application of image super-resolution reconstruction technology. Thisdissertation had comparatively done systematic study to super-resolutionreconstruction and the application of sparse representation in imagesuper-resolution reconstruction. Some of the existing algorithms were improved.Firstly, this dissertation had done study to the common image super-resolutionreconstruction algorithms and the sparse representation theoretical. For the imagesuper-resolution reconstruction algorithms, these algorithms which divided into twotypes (based on reconstruction and based on learning) were introduced in details,and the advantages and disadvantages of these algorithms were analysed. For sparserepresentation theory, this dissertation focused on the solving of ill-posed problemand the designing of over-complete dictionary, and applied these to the imagesuper-resolution reconstruction.Secondly, this dissertation had done study to the structure of over-completedictionary and non-local similarity theory based on classic sparse, and presented anew method of based on clustering for single-image super-resolution reconstruction. First,a high-resolution dictionary was learned from the samples of high-resolutionimages; second, got the representation coefficients of high-resolution image byusing iterative shrinkage solution; finally, used the high-resolution dictionary toreconstruct the low-resolution image.Finally, this dissertation compared the method to the classic methods within noiseenvironment and without noise environment. The results of the simulate experiments showthat this method can effectively improve the effect of the reconstruction and reduce thecomplexity of the algorithm.更多还原