摘要
考虑目前无需亚像素精度配准的三维导向核回归(3D-SKR)超分辨率重建算法对图像中的离群点高度敏感的问题,引入了稳健估计中的Huber函数,并结合中值滤波,提出了一种稳健的三维导向核回归超分辨率重建算法。该算法将图像中残差大于Huber尺度参数的点视为离群点,利用三维中值滤波对其进行隐藏,然后再使用Huber函数进行超分辨率重建。实验证明,该算法在保持了原有算法的良好特性的基础上,有效地消除了离群点对重建结果的影响。
Considering the issue of the 3D-SKR super resolution reconstruction which doesn＇t need sub-pixel registration being highly sensitive to the outliners in the image, we proposed a robust scheme of 3D-SKR super resolution reconstruction by introducing the Huber function in M-estimators integrated with the median filtering. This scheme regards the points whose residual error is greater than the scale parameter of Huber as the outliners, which would be concealed by 3d median filtering, and then the super resolution reconstruction is carried out with Huber function. Experiment results proved that improved algorithm can remove the influence caused by the outliners with the maintaining of the good performance of the original algorithm.
出处
《电子测量技术》
2012年第6期84-87,共4页
Electronic Measurement Technology
关键词
超分辨率
三维导向核回归
序列图像
Huber函数
中值滤波
super-resolution
3d-steering kernel regression（3D-SKR）
image sequences
Huber function
median filtering
