多帧图像超分辨复原算法中帧的选取  

Frames Selection in Multi-frame Image Super-resolution Restoration

作　　者：张志[1,2] 王润生[1] 

机构地区：[1]国防科技大学ATR实验室,湖南长沙410073 [2]空军装备研究院,北京100085

出　　处：《信号处理》2009年第11期1775-1780,共6页Journal of Signal Processing

基　　金：国家自然科学基金资助项目(60302007);国防科技重点实验室基金项目(9140C8004010611;9140C8002010705)

摘　　要：研究了多帧图像超分辨复原算法中参考帧和参与复原图像帧数据集的选取问题。提出了两条参考帧选取的准则,设计了一种分步的图像配准和超分辨复原的方法。首先计算初始图像序列的运动向量参数,根据准则选取了参考帧和参与复原图像帧数据集,进而生成了新的有利于图像超分辨复原的优化的图像序列,再通过亚像素图像配准,计算得到小范围的相对精确的运动估计参数,最后用到图像超分辨复原算法中。仿真实验结果证明,本文提出的多帧图像超分辨复原算法中帧的选取方案是合理的,超分辨复原的图像质量在主观评价和客观指标上都获得了优于经典的Farsiu算法的效果。Selection of reference frame and the considering frames is investigated for multi-frame image super-resolution restoration. Two rules of reference frame selection are proposed. It devises a framework to resolve the image super-resolution restoration problem into two steps. Firstly, the motion vector is obtained by image registration method. The selection of reference frame and the considering frames is obtained according to the rules, and then the refined image sequence is generated. Secondly, the accurate motion vector is calculated. At last, the high-resolution image is generated under the framework of classical image super-resolution. Experimental results indicate that the proposed method is better than Farsiu＇ s method both in subjective and objective image quality assessment.

关 键 词：图像复原 超分辨 参考帧 图像配准 运动估计 光流法 

分 类 号：TP911.73[自动化与计算机技术]