摘要
本文提出了一种基于图像块自适应融合的序列图像超分辨率重建算法。算法使用低分辨率序列图像中的互补信息重建生成高分辨率图像。为了保证重建初始估计尽可能接近真实场景,配准后的序列图像按照图像块梯度信息自适应的融合生成高分辨率初始估计图像。算法采用误差反向投影的方法对高分辨率图像迭代校正,生成超分辨率重建最终结果。实验证明,本文提出的超分辨率重建算法能够在增加图像细节信息的同时重建出更加自然真实的高分辨率图像。
In this paper,we propose a super-resolution reconstruction algorithm based on adaptive patch fusion. This algorithm well fulfills its intention to reconstruct high-resolution image from observed low-resolution image sequence. The observed low-resolution images are aligned by SURF feature-point registration in the first place,and are adaptively fused into a single high-resolution image based on gradient prior subsequently. Image fusion process is conducted in a patch-wise manner which ensures that utmost local prior-detail is incorporated in the initial guess of high-resolution image. Our algorithm further incorporates an iterative error back projection mechanism to optimize the high-resolution image estimation. Experimentsshow that our patch based adaptive fusion mechanism enhances image details naturally and reconstructs high resolution images of space objects better than state-of-the-art methods.
出处
《中国体视学与图像分析》
2015年第2期99-107,共9页
Chinese Journal of Stereology and Image Analysis
基金
国家自然科学基金(编号61071137
61371134)
关键词
超分辨率重建
序列图像配准
图像梯度
图像块融合
super-resolution reconstruction
sequential-image registration
image gradient
patch fusion