一种基于子像素配准视频超分辨率重建方法

A video super-resolution reconstruction method based on sub-pixel registration

为了提高视频的空间分辨率,提出了一种利用帧间运动信息进行超分辨率重建的方法。对于整个视频的重建,提出了一种基于滑动窗的分段重建模型。在每一个滑动窗中,首先对相邻帧进行子像素级精度的运动配准;然后通过迭代反投影算法进行超分辨率重建。在配准算法中,提出了一种基于四参数刚体变换模型的配准方法,通过迭代求解和高斯金字塔图像模型由粗及精地进行运动估计。分别对模拟图像及实拍彩色视频进行重建,实验结果表明,该配准算法具有较高的精度,重建算法取得了较高的峰值信噪比(PSNR)值,重建视频具有更好的视觉效果和更高的分辨率能力,可被广泛应用于在帧间主要存在平移和旋转运动的视频序列的超分辨率重建。

To improve the spatial resolution of a video,a super-resolution reconstruction method is proposed utilizing the movement information between frames.For the reconstruction of the whole video,a segmentation reconstruction model based on sliding window is proposed.In each sliding window,firstly,a movement registration with sub-pixel precision is performed on the adjacent frames;Secondly,the super-resolution reconstruction is implemented using the iterative back projection algorithm.In the registration algorithm,a method based on four parameter rigid transformation model is proposed.The iterative solving method and the Gauss pyramid image model are used to estimate the movement parameters from coarseness to fine.The reconstruction is performed on simulated image and real-taken color video respectively.The experimental results show that this registration algoritm gets better precision,and the reconstruction algorithm achieves higher PSNR,and the reconstructed video has better visual effect and higher resolution ability.It can be extensively applied to the super-resolution reconstruction of video sequences in which the frames are different from each other mainly by translation and rotation.