【作者】 刘雅闻；

【导师】 门爱东；

【作者基本信息】 北京邮电大学 ， 通信与信息系统， 2009， 硕士

【摘要】 在图像处理领域中,图像的超分辨率重建技术和图像压缩是两个发展的热点问题。本文从实际应用的要求出发,对二者的结合作了研究,即对压缩图像进行超分辨率重建。论文的主要工作如下:1)针对超分辨率技术在压缩视频上的特殊应用,本文综合分析了压缩算法所引入的量化噪声特性以及多帧图像超分辨率算法的数学原理,并在考虑降质模型的不确定性以及反卷积求解方法固有的病态性的基础上,提出了一种较低复杂度的且有效可行的空域重建框架。2)对超分辨率重建技术中的配准算法进行研究。配准的准确度是超分辨率重建算法成功与否的关键和基础,它要求配准算法中的运动估计必须是亚像素级的。本文在提出基于光流场配准方法的基础上,结合高斯金字塔分层结构,实现了由粗到精的亚像素运动估计。3)针对LK光流法跟踪效果易受噪声影响的现象,本文研究了压缩视频量化噪声(块效应、振铃效应)的特性,提出了一种基于区域检测的预处理算法来减弱该噪声对配准结果的影响。4)对超分辨率重建中插值的理论基础及典型算法进行研究。本文综合考虑了运算复杂度及插值效果,将Peyman Milanfar的核回归(kernel Regression)思想应用于超分辨率的插值重建算法中,实现了对配准后数据的较好拟合。更多还原

【Abstract】 In the field of image processing, the technology of super resolution reconstruction and image compression are popularly developed. In order to meet the increasing demand of reconstruction one or a sequence of high resolution image from compressed video, this paper focuses on following aspects:1) For working on the special application of super resolution technique on compressed video data reconstruction, we have studied the new questions introduce by compression algorithm and the theory of the muti-frame image super resolution. Considering degradation model can not be built exactly, we propose an effective spacial reconstruction framework of low complexity.2) In this paper, we have made some research on image registration algorithm in super resolution reconstruction. Since the accuracy of the registration algorithm results play an very important role in super resolution reconstruction, we propoded an algorithm that combines the optical flow method and Gaussian Pyramid method and achieve a motion estimation of subpixel accuracy.3) Since the track effect of the LK optical flow method can be badly affected by noise, this paper proposes a preprocessing algorithm, which is based on region detection, to weaken this affection, after studing the properties of the quantization noise in compressed video.4) In this paper, we study the basic theory and typic algorithms of the interpolation in super resolution reconstruction. And apply the kernel Regression theroy by Peyman Milanfar on interpolation algorithm of super resolution reconstruction by balancing the caculation complexcity and the effection. And this algorithm proposed can make good simulation of the data after registration.更多还原