提出用粒子滤波器来实现视频序列的稳像,解决了视频序列的帧间不稳定问题。提出的稳像算法从视频图像中提取角点特征,建立当前帧与参考帧之间的映射关系,然后根据仿射变换模型求取最小二乘解来获得帧间的全局运动参数,最后利用粒子滤波平滑运动参数,实现帧间的实时运动补偿。对包含80帧场景的视频序列进行了实验,稳像后视频序列的平均峰值信噪比比稳像前提高了24.88,同时稳像精度＜1 pixel,处理时间＜30 ms。实验结果表明,本文算法能有效地改善图像质量,在去除高频抖动的同时能较好地保留摄像机的主动运动,稳像效果良好。

A digital image stabilization algorithm based on a particle filter is present to remove the interframe vibration of video images. Firstly,the corners of video images are extracted using Harris operators,then the mapping relationship between the current frame and the reference frame is established and global motion vectors are obtained by computing least-square solution based on an affine transformation model. Finally,the particle filtering is used to smooth motion vectors to realize the motion compensation of video frames. Experiments are undertaken for a video sequence with 80 frames,it could be found that the average Peak Signal-to-Noise Ratio (PSNR) of the image sequence after image stabilization is 24.88 higher than that of the original video sequence while providing the image stabilization accuracy and the processing time low than 1 pixel and 30 ms respectively.These results show the proposed stabilization algorithm improves image quality effectively,and it can not only alleviate the vibration but also can preserve the initiative motion of the camera.