【作者】 何宇清；

【导师】 侯正信；

【作者基本信息】 天津大学 ， 信号与信息处理， 2007， 博士

【摘要】 图像超分辨率处理技术是指利用多帧关于同一场景的有相互位移的低分辨率降质图像来重建高分辨率高质量图像的技术。图像超分辨率处理技术可突破图像采集设备的分辨率限制,充分利用多帧图像之间的互补信息,实现像素级的图像信息融合。图像超分辨率处理和成像技术在遥感、军事、高清晰度电视、医学成像、公共安全等领域有很广泛的应用前景。图像超分辨率重建算法可以分为两大类,即频域算法和空间域算法。频域算法只能对全局位移的图像序列进行处理;空间域算法使用通用的观察模型,具有更好的适应性和重建效果,是目前的主要研究方向。空间域算法包括迭代反投影方法,最大后验概率方法(MAP),凸集投影法(POCS),混合MAP/POCS方法以及自适应滤波方法等。本文全面介绍了图像超分辨率重建的研究现状,针对参考帧重建,运动估计、利用运动估计信息共轭梯度法加权重建图像,重建图像的质量评价三个方面作了深入地研究。本文探讨了传统的数字图像内插算法,研究了使用全相位内插算法来重建参考帧,并根据数字图像在边缘轮廓局部特点:跨越边缘方向的灰度值变化比较尖锐,沿边缘方向的灰度值变化比较平滑;提出了改进的全相位内插算法来重建参考帧。运动估计是超分辨率图像重建的一项核心技术,在图像重建过程中的作用是把观察帧投影到参考帧上,运动估计的准确性对重建效果有决定性的影响,关键是如何准确地获得半像素运动信息。本文提出了基于帧差的运动估计方法,该方法与传统块匹配运动估计(BMA)相比,搜索结果准确,尤其提高了边缘区域的搜索准确性,符合图像重建算法的要求。块匹配运动估计方法和光流法运动估计方法各有优缺点,本文将二者相结合加权,用共轭梯度法进行图像超分辨率重建。超分辨率重建图像的评价,目前没有统一的方法。本文针对超分辨率重建图像的特点,提出了基于边缘基元和边缘局部相似度的评价方法,边缘区域是图像超分辨率重建的重点,对人眼观察图像起到重要作用,本文研究了考虑人眼观看感受,加强边缘的超分辨率重建方法。同时,文中对各算法做了仿真实验,得到了较理想的结果。更多还原

【Abstract】 Image super resolution (SR) processing is the technology to reconstruct high resolution and high quality images from a group of warped, blurred and noised low resolution (LR) images about the same scene. It breaks through the resolution limit of image acquisition equipment, uses complement information in multi-frame and can achieve data fusion on pixel level. Image SR processing has proved to be useful in many practical applications, such as remote sensing, military detection, HDTV, medical imaging, machine vision and public security, etc.Super resolution algorithms include frequency-domain approach and spatial-domain approach. Frequency domain approach can only deal with image sequences that only translational motions are allowed. Spatial methods which are the major research directions have better adaptability and performance by using general observation models. Spatial methods include the iterative back-projection (IBP) method, stochastic method (MAP), projection on convex set (POCS) method, hybrid reconstruction method and adaptive filter method, etc. This paper introduces research of image super resolution reconstruction technologies all-around and studies three areas in-depth: reference frame reconstruction; motion estimation and conjugate gradient image reconstruction that used motion-estimated information; quality assessment of image reconstruction.Traditional image interpolation algorithms are study in this paper. For reference frame reconstruction, all-phase interpolation algorithms are put forward. Base on two factors of an edge profile: the gray value across the edge orientation is sharp and the gray value along the edge orientation is smooth, this paper proposes modified all-phase interpolation algorithm.Motion estimation is an important technique in super resolution problems, in SR reconstruction motion estimation is used to project the observation frames onto the reference frame. The accuracy of motion estimation has great effect to the reconstruction results. The key to the SR image reconstruction is the accurate knowledge of the sub-pixel motion information of the neighbor frames, this paper put forward motion estimation algorithm base on frame differences. Compared with traditional block matching algorithm (BMA), the results of this algorithm are accuracy especially in edge area. Considering the benefit and defect of BMA and optical flow method, this paper studies combined method, gets weight information, and uses conjugate gradient to reconstruct SR image.At present, there’s no uniform assessment method on the quality of SR reconstruction image. Considering the characteristics of SR image, the assessment method based on edge block and the relative of LR image edge is proposed. Edge areas are important in SR reconstruction, especially for human being’s visual system. Edge emphasized method is studied based on the human visual sensation. In this paper, these algorithms are emulated and the results are better.更多还原