计算机工程与应用 ›› 2012, Vol. 48 ›› Issue (28): 206-213.

• 图形、图像、模式识别 • 上一篇    下一篇

基于多帧融合及四参数仿射模型的图像超分辨

赵洪达1,2,刘本永1,2   

  1. 1.贵州大学 计算机科学与信息学院,贵阳 550025
    2.贵州大学 智能信息处理研究所,贵阳 550025
  • 出版日期:2012-10-01 发布日期:2012-09-29

Image superresolution based on multiframe processing and 4-parameter affine model

ZHAO Hongda1,2, LIU Benyong1,2   

  1. 1.College of Computer Science and Information, Guizhou University, Guiyang 550025, China
    2.Institute of Intelligent Information Processing, Guizhou University, Guiyang 550025, China
  • Online:2012-10-01 Published:2012-09-29

摘要: 运动参数估计和复原是多帧图像超分辨重构中最重要的两个环节,其中经典的Fourier-Mellin变换方法于频域采用对数极坐标形式和相位相关方法结合来估计运动参数。相位相关是整像素级平移参数估计方法,将其改进为亚像素级平移参数估计方法,以提高旋转、缩放参数的估计精度。对于复原算法,在讨论基于局部信息的传统双三次插值超分辨重构方法的基础上,重点探讨基于全局信息的Kriging插值超分辨重构和核非线性回归(KNR)超分辨重构方法。实验结果表明,探讨的参数估计方法和超分辨重构方法是有效的。

关键词: 图像超分辨, 四参数仿射模型, Fourier-Mellin变换, Kriging插值, 核非线性回归(KNR)

Abstract: Motion estimation and reconstruction are the two most important steps in image superresolution reconstruction based on multiframe processing. For this purpose, Fourier-Mellin transform algorithm is one of the most popular approaches which combines the log-polar image coordinate in the frequency spectrum with phase correlation to estimate the motion parameter. However, the estimation precision of phase correlation is limited to pixel level, and in this manuscript it is improved to sub-pixel level to increase the estimation accuracy of rotation and scaling parameter. As for reconstruction algorithm, in comparison with the standard bicubic interpolation which is based on the local information contained in an image, the Kriging interpolation and Kernel Nonlinear Regression(KNR) superresolution algorithms are discussed which are based on the global information. The effectiveness of the motion estimation and the superresolution reconstruction algorithms discussed in the paper is illustrated by some experimental results.

Key words: image superresolution, 4-parameter affine model, Fourier-Mellin transform, Kriging interpolation, Kernel Nonlinear Regression(KNR)