基于低层次计算机视觉的超分辨率图像重建

Low-level vision based super-resolution image reconstruction

在基于低层次计算机视觉的超分辨率图像重建过程中,角点检测和插值是两个关键的技术。首先在SUSAN角点检测算法的基础上提出了改进算法,改进后的算法根据图块对比度的不同,在确定位于不同图块中的像素的USAN面积时采用了可变灰度阈值,可变灰度阈值的采用,使得检测出的角点分布更加均匀,而角点分布均匀则使得图像配准更加精确,有利于后期的重建工作。其次,提出了一种适合于超分辨率图像重建的插值算法:基于圆区域的自适应插值算法。该算法可以根据待插值点周围的灰度特征自适应决定插值策略,将线性插值、最邻近插值和中值插值法有机地结合在一起。大量的仿真实验证明了提出算法具有运算量小、图像重建后的效果出重,易于实现。

In the process of super-resolution image reconstruction,corner detection and interpolation are two key technologies.In this paper,two approaches are proposed respectively.Firstly,an improved corner detection approach based on SUSAN principle for feature detection is proposed.It has variable thresholds for the allowed variation in brightness within the USAN.It makes corner well-distributed.Secondly,a new adaptive interpolation approach based on circular-area is proposed.The approach can adaptively seleet the interpolation method based on gray feature around the interpolated point.In this method,linear interpolation,nearestneighbor interpolation and median interpolation are merged.Experiments confirm that these algorithms are superior to others.