华南理工大学学报(自然科学版) ›› 2011, Vol. 39 ›› Issue (5): 84-90.doi: 10.3969/j.issn.1000-565X.2011.05.015

• 计算机科学与技术 • 上一篇    下一篇

基于Keren配准和插值的快速超分辨率重建

李展1 韩国强1 陈湘骥2 廖秀秀1   

  1. 1.华南理工大学 计算机科学与工程学院,广东 广州 510006; 2.华南农业大学 信息学院,广东 广州 510642
  • 收稿日期:2010-09-17 修回日期:2011-01-04 出版日期:2011-05-25 发布日期:2011-04-01
  • 通信作者: 李展(1979-) ,女,博士生,讲师,主要从事超分辨率图像重建、图像恢复、天文图像处理研究. E-mail:lizhan@jnu.edu.cn
  • 作者简介:李展(1979-) ,女,博士生,讲师,主要从事超分辨率图像重建、图像恢复、天文图像处理研究.
  • 基金资助:

    国家自然科学基金资助项目( 10778617,10973007,61070090) ; 广东省科技计划重大专项( 2010A080402005) ;广东省科技计划项目( 2008B080701052,2010B080701062) ; 广东省自然科学基金资助项目( 10151063201000002)

Fast Super-Resolution Image Reconstruction Based on Keren Registration and Interpolation

Li ZhanHan Guo-qiangChen Xiang-jiLiao Xiu-xiu1   

  1. 1. School of Computer Science and Engineering,South China University of Technology,Guangzhou 510006,Guangdong,China; 2. College of Informatics,South China Agricultural University,Guangzhou 510642,Guangdong,China
  • Received:2010-09-17 Revised:2011-01-04 Online:2011-05-25 Published:2011-04-01
  • Contact: 李展(1979-) ,女,博士生,讲师,主要从事超分辨率图像重建、图像恢复、天文图像处理研究. E-mail:lizhan@jnu.edu.cn
  • About author:李展(1979-) ,女,博士生,讲师,主要从事超分辨率图像重建、图像恢复、天文图像处理研究.
  • Supported by:

    国家自然科学基金资助项目( 10778617,10973007,61070090) ; 广东省科技计划重大专项( 2010A080402005) ;广东省科技计划项目( 2008B080701052,2010B080701062) ; 广东省自然科学基金资助项目( 10151063201000002)

摘要: 为提高图像超分辨率重建技术实时应用的可能性,增强其对配准误差的容忍度,提出了一种基于Keren 配准和插值的快速鲁棒超分辨率图像重建算法.该算法将配准后的低分辨率图像根据变换参数映射到高分辨率网格上,再利用模板卷积迭代地填充缺失像素值,从而重建一幅高分辨率图像.将文中算法与非均匀插值法、凸集映射法、鲁棒的迭代后向映射法和结构适应的归一化卷积法4 种超分辨率图像重建算法进行了比较.实验结果表明,文中算法对一定精度范围内的配准误差不敏感,在速度和重建效果上具有一定的优势,是一种有效、鲁棒和快速的多帧超分辨率图像重建算法.

关键词: 图像重建, 超分辨率, Keren 配准, 插值, 卷积

Abstract:

In order to make it possible to apply the super-resolution reconstruction ( SRR) technology of images in real time and to improve the tolerance of registration errors,a fast and robust SRR algorithm based on Keren registration and interpolation is proposed. In this algorithm,registered low-resolution ( LR) images are mapped onto a high-resolution ( HR) grid according to their transform parameters,and the space pixels are filled iteratively via the template convolution to reconstruct a HR image. The proposed algorithm is finally compared with four existing SRR algorithms including the nonuniform interpolation,the projection onto convex sets,the robust iterative back projection and the structure-adaptive normalized convolution. The results show that the proposed algorithm is an effective,robust and fast SRR method for multi-frame images because it is insensitive to registration errors in a certain accuracy range with high reconstruction speed and quality.

Key words: image reconstruction, super resolution, Keren registration, interpolation, convolution